Hydroclimatic Variability in the Bilate Watershed, Ethiopia

: It is important to understand variations in hydro-meteorological variables to provide crucial information for water resource management and agricultural operation. This study aims to provide comprehensive investigations of hydroclimatic variability in the Bilate watershed for the period 1986 to 2015. Coefﬁcient of variation (CV) and the standardized anomaly index (SAI) were used to assess the variability of rainfall, temperature, and streamﬂow. Changing point detection, the Mann–Kendell test, and the Sen’s slope estimator were employed to detect shifting points and trends, respectively. Rainfall and streamﬂow exhibited higher variability in the Bega (dry) and Belg (minor rainy) seasons than in the Kiremt (main rainy) season. Temperature showed an upward shift of 0.91 ◦ C in the early 1990s. Reduction in rainfall ( − 11%) and streamﬂow ( − 42%) were found after changing points around late 1990s and 2000s, respectively. The changing points detected were likely related to the ENSO episodes. The trend test indicated a signiﬁcant rise in temperature with a faster increase in the minimum temperature (0.06 ◦ C/year) than the maximum temperature (0.02 ◦ C/year). Both annual mean rainfall and streamﬂow showed signiﬁcant decreasing trends of 8.32 mm/year and 3.64 mm/year, respectively. With signiﬁcant increase in temperature and reduction in rainfall, the watershed has been experiencing a decline in streamﬂow and a shortage of available water. Adaptation measures should be developed by taking the increasing temperature and the declining and erratic nature of rainfall into consideration for water management and agricultural activities.


Introduction
Climate change has been influencing hydroclimatic systems and caused various impacts with different magnitude and frequency of hydrological extremes to water resources management, agricultural activities [1][2][3], and economic development [4]. Although warming and cooling are distinct in temporal and spatial regions, substantial warming was observed in Sub-Saharan African countries [5]. Millions of people in Africa are affected by droughts and floods due to hydroclimatic variability. Drought impacts to developing countries with low adaptive capacities are often acute in crop failures, hunger, and loss of life and property. In the mid-1980s, the horn of Africa suffered a shortage of water, which led to economic losses of several hundred million dollars [6]. Rainfall deficits in 1984Rainfall deficits in , 2009, and 2015 were among the most severe droughts that impacted agricultural production and human life in Ethiopia [7]. On the other hand, extreme floods in 1988, 1996, and 2006 caused massive loss of life and properties in the country [8]. For example, an overflow from the Bilate River at the end of July 2006 affected 5370 households in the Humbo Woreda [9,10]. This flood damage has been substantial and exceptional in the watershed.
The Ethiopian economy mainly depends on agriculture, which contributes approximately 43% of total gross domestic product (GDP) and generates 90% foreign exchange earnings [11]. However, the dependence of the country's economy on rainfed practices has caused agricultural productions to be highly vulnerable to climate hazards such as floods 1). The watershed has three seasons classified based on rainfall amount. They are called Kiremt/main rainy season (June-September), Bega/dry season (October-January), and Belg/minor rainy season (February-May). Rainfall distribution shows a bimodal pattern with peaks in Kiremt and Belg. These seasons determine the agricultural activities of farmers [36]. Seasonal rainfall in the area is highly variable and difficult to predict, while the temperature is slightly seasonal [26].
Mean annual rainfall in the Bilate watershed ranges from 769 mm in the lowlands to 1339 mm in the highlands. The mean annual temperature varies from 11 °C to 22 °C at Hossana, the upper stream, and 16 °C to 30 °C at Bilate Tena, the watershed's lower stream. The daily river flows are between 1.2 and 31.25 m 3 /s at Alaba Kulito and 3 to 45.7 m 3 /s at Bilate gauging station in the upper and lower streams, respectively [35].
The Bilate watershed is characterized by mixed cultivation and supports irrigation to grow commercial plants such as tobacco, vegetables, and maize. Demand for water irrigation is increasing, and medium-sized private investors have been accessing below the required water demand [37].

Datasets
Observed daily rainfall from twelve stations and temperature from nine stations during 1986 to 2015 were collected from the Ethiopian National Meteorological Agency (NMA). The daily streamflow data of the Bilate Tena gauging station from 1986-2015 and 30 m by 30 m digital elevation model (D.E.M.) were collected from the Ministry of Water, Irrigation, and Electricity of Ethiopia. The DEM was used to delineate the study area.

Methods
This study used different variability and trend analysis techniques to analyze the hydroclimatic time series of streamflow data at the Bilate gauging station, basin-averaged Mean annual rainfall in the Bilate watershed ranges from 769 mm in the lowlands to 1339 mm in the highlands. The mean annual temperature varies from 11 • C to 22 • C at Hossana, the upper stream, and 16 • C to 30 • C at Bilate Tena, the watershed's lower stream. The daily river flows are between 1.2 and 31.25 m 3 /s at Alaba Kulito and 3 to 45.7 m 3 /s at Bilate gauging station in the upper and lower streams, respectively [35].
The Bilate watershed is characterized by mixed cultivation and supports irrigation to grow commercial plants such as tobacco, vegetables, and maize. Demand for water irrigation is increasing, and medium-sized private investors have been accessing below the required water demand [37].

Datasets
Observed daily rainfall from twelve stations and temperature from nine stations during 1986 to 2015 were collected from the Ethiopian National Meteorological Agency (NMA). The daily streamflow data of the Bilate Tena gauging station from 1986-2015 and 30 m by 30 m digital elevation model (D.E.M.) were collected from the Ministry of Water, Irrigation, and Electricity of Ethiopia. The DEM was used to delineate the study area.

Methods
This study used different variability and trend analysis techniques to analyze the hydroclimatic time series of streamflow data at the Bilate gauging station, basin-averaged rainfall from twelve stations, and basin-averaged temperature from nine stations. Variability analysis involves using a standardized anomaly index (SAI) [38] and the coefficient of variation (CV) [39], while trends and abrupt changes detection using the Mann-Kendall (MK) test [40,41], Sen's slope estimator [42], and three change-point detection methods. The spatial distribution of annual and seasonal precipitation was presented by the inverse distance weighted interpolation (IDW) [43]. It is suitable for the interpolation of highly variable climatic data. It also assumed that the non-sampled point attribute's value is a weighted average of the known values in the vicinity and that weights are inversely proportional to the distance between the prediction site and sampled location [44]. The leave one out cross-validation (LOOCV) method [45], which is done by removing one data point at a time and using the remaining data to interpolate for the location of the removed point, can be applied for IDW validation. The performance of the IDW can be further evaluated by the Pearson's correlation coefficient [46] and the root mean square error (RMSE) methods [47]. In this study, the IDW method was selected to present areal rainfall estimation. Hydrometeorological variabilities were mainly conducted by using basin-averaged rainfall and temperature data.

Variability Analysis
The variability of rainfall, temperature, and streamflow was analyzed using the coefficient of variation (CV) and standardized anomaly index (SAI).

Coefficient of Variation (CV)
The coefficient of variation measures the total deviation from the mean value. It is calculated as the ratio between the standard deviation and the long-term data mean, multiplied by 100% [48]. The CV is used to compare long-term variations of hydroclimatic variables in seasonal and annual scales [49].
where CV is the coefficient of variation; µ is the mean hydroclimatic variable; and δ is the standard deviation over the period. Degree of variability was classified as less (CV < 20), moderate (20 < CV < 30), and high (CV > 30). In general, a higher CV shows a more considerable variation in hydroclimatic variables and vice versa [50].

Standardized Anomaly Index (SAI)
The standardized anomaly index helps to identify dry and wet years with long-term mean rainfall and examines the nature of trends [1,50]. It is obtained by calculating the difference between the annual total and the mean long-term rainfall for a given year divided by the long-term data's standard deviation. The seasonal scale rainfall anomaly can also be calculated using this index, which is expressed as: where SAI = standardized rainfall anomaly; x = annual rainfall in year t; µ = long-term annual mean rainfall over a given period of observation; and δ = standard deviation of rainfall data. According to classifications given in [51], values of SAI are classified as shown in Table 1.

Changing Point Detection Test
Studying whether or not there are sudden changes in time series and identifying the shift is essential for understanding a river basin's sensitivity to climate variability and impacts on water resource availability [52]. We utilized Buishand's range (BR) [53], standard normal homogeneity (SNH) [54], and Pettitt's [55] changing point detection methods to investigate the presence of sharp shifts in annual and seasonal time series of hydroclimatic variables in the Bilate watershed. A non-parametric Pettitt test detects a significant change in the time series average if the change's exact time is unknown. This test has been widely used to identify observed changes in climatic and hydrological datasets. The Pettitt test is more sensitive to breaks in the middle of the time series [56,57]. Under a null hypothesis, this test assumes that there is no changing point and the alternative hypothesis considers a date at which there is a change in the data. If the p-value is less than the 0.05 significance level, the null hypothesis is rejected to accept the alternative hypothesis [21].
The standard normal homogeneity test (SNHT), unlike the Pettitt test, is more sensitive to breaks that appeared at the beginning and end of the data series. The null hypotheses and alternative hypothesis assumptions of the SNH test are the same as those of the Pettitt test. This method detects a change in a time series of data by comparing the mean of the record's first observations with the last to form the test statistic [22]. According to the test statistics' critical values for different detection tests listed in Table 2 in [58], the changing point based on the total number of observations may be the sample where the test statistics reach the maximum value. When the critical value is less than the test statistic, the null hypothesis can be rejected, depending on the sample size.
The Buishand range (BR) method detects the changing point based on cumulative deviations from the mean. This method assumes that the testing variable's values have the same mean, and under the alternate hypothesis, it takes the existence of possible changes in the mean value [53]. Overall, a series may be considered as having a changing point or be inhomogeneous if at least two tests reject the null hypothesis; and considered homogeneous or no changing point if no test out of three tests rejects the null hypothesis.
After detecting the changing point of rainfall and streamflow, we further examined the stationarity of precipitation and streamflow before and after the changing point using frequency analysis. We calculated the exceedance probability of rainfall and streamflow, which were plotted using daily data, defined the rainfall (R i ) and discharges (Q i ) arranged in descending order, where R 1 and Q 1 being the largest values of rainfall and streamflow, respectively. The exceedance probability was calculated as [59]: where n represents the number of days of rainfall and streamflow, and i indicates the rank.

Trend Analysis
Detection of trends in hydroclimatic variables can be performed using parametric and non-parametric tests. Parametric tests are robust but require the available data for normal distribution, stationarity, and serial independence [60]. It is rarely used for climatic and hydrological data because some are biased and contain outlier observations [61]. Nonparametric statistical analyses are not affected by the normal distribution of data, are less sensitive to outliers, and are better suited to normally undistributed data [1,61,62]. This characteristic makes a non-parametric test for the trend detection of hydroclimatic variables advantageous over a parametric analysis. We applied the Mann-Kendall (MK) test [40,41], which is one of the most widely known non-parametric trend detection tests used for hydroclimatic time series [63]. The expressions for the Mann-Kendall test statistics (S) is given by: where n is the number of the dataset; x j and x i are consecutive data values on years j and i, where j > i; and sgn (x j − x i ) is computed using Equation (5).
The variance, Var(s) is calculated by: (6) where n denotes the number of observations; m represents the number of tied groups; and t i indicates the number of observations in the i-th data group. The standard test statistic Zs was computed using: The test was used for hydroclimatic variables on a seasonal and annual basis with a confidence level of 95%. According to this test, the null hypothesis (H 0 ) designates that there is no trend in the series, while the alternative hypothesis (H 1 ) describes that there is a trend [64]. Test statistic Zs evaluates the statistical significance of the trend. If Zs is positive, it indicates an upward trend while negative denotes decreasing trends [65]. The trend is reflected as significant at a significance level of 0.05 when |Z s | > 1.96 [66]. In this study, a 0.05 significance level for the analysis of the hypothesis was considered.
A non-parametric Sen's slope estimator [42] was used to estimate the time series trends' magnitude and direction. The slope (β) is calculated using the equation given by.
where β represents the median of the slope values between data points x i and x j at time step i and j, respectively (i < j). The value of β indicates the magnitude of the trend, the sign of β specifies the direction of the data trend, a positive value of β shows an upward trend, and a negative implies a downward trend. The Sen slope estimation method is more robust than linear regression, which reduces the outliers' effect or missing values and performs better, even for normally distributed data [67].

Calculation of Potential Evapotranspiration (PET) and Evapotranspiration (ET)
In this study, the daily potential evapotranspiration (PET) was calculated using the Hargreaves method (Equation (9)), which uses the minimum and maximum temperatures as input. In addition, we calculated evapotranspiration (ET) with the Budyko-type relation (Equation (10)) [68,69] and the basin water storage (∆S) with the water balance equation [70].
where PET denotes the Hargreaves potential evapotranspiration (mm/day); R a represents extraterrestrial radiation (mm/day) that depend on latitude, sunshine hours, and solar constant; λ is the latent heat of vaporization (MJ/Kg); and T min and T max indicate the minimum and maximum temperature ( • C).
where w is a parameter that depends on the watershed characteristics. For tropical climates, Pike proposed w = 2 [71].

Climatology of Study Area
Monthly mean maximum temperature increases from August (23.81 • C) and peaks in February (27.86 • C), then gradually decreasing and reaching the lowest point in July (23.48 • C) were observed in the Bilate watershed, as shown in Figure 2. With the increase in elevation (Figure 1), the temperature dropped from the south toward the north of the watershed. Substantial amounts of rainfall were mainly observed during the Kiremt (49.4%, June-September) and the Belg (36.5%, February-May) seasons, as shown in Figure 2. Streamflows of the Kiremt, Belg, and Bega seasons account for 53.6%, 27.4%, and 19.0%, respectively, of annual flow. tion (Equation (10)) [68,69] and the basin water storage (ΔS) with the water balance equation [70].
where PET denotes the Hargreaves potential evapotranspiration (mm/day); Ra represents extraterrestrial radiation (mm/day) that depend on latitude, sunshine hours, and solar constant; λ is the latent heat of vaporization (MJ/Kg); and Tmin and Tmax indicate the minimum and maximum temperature (°C).
where w is a parameter that depends on the watershed characteristics. For tropical climates, Pike proposed w = 2 [71].

Climatology of Study Area
Monthly mean maximum temperature increases from August (23.81 °C) and peaks in February (27.86 °C), then gradually decreasing and reaching the lowest point in July (23.48 °C) were observed in the Bilate watershed, as shown in Figure 2. With the increase in elevation (Figure 1), the temperature dropped from the south toward the north of the watershed. Substantial amounts of rainfall were mainly observed during the Kiremt (49.4%, June-September) and the Belg (36.5%, February-May) seasons, as shown in Figure  2. Streamflows of the Kiremt, Belg, and Bega seasons account for 53.6%, 27.4%, and 19.0%, respectively, of annual flow. The complex topographical features play an essential role in Ethiopia's climate [14]. The country has a diverse topography that ranges from 125 m below sea level at the Danakil Depression to 4620 m above sea level at RasDejen. The country is also dominated by a plateau with a mountain range separated by the East African Rift Valley [72]. These physiographic variations create a significant difference in climatological condition both spatially and temporally. Subsequently, the mean annual rainfall of Ethiopia is characterized by 141 mm in the lowlands and 2275 mm in the highlands [16]. The complex topographical features play an essential role in Ethiopia's climate [14]. The country has a diverse topography that ranges from 125 m below sea level at the Danakil Depression to 4620 m above sea level at RasDejen. The country is also dominated by a plateau with a mountain range separated by the East African Rift Valley [72]. These physiographic variations create a significant difference in climatological condition both spatially and temporally. Subsequently, the mean annual rainfall of Ethiopia is characterized by 141 mm in the lowlands and 2275 mm in the highlands [16].
The spatial distribution of annual and seasonal precipitation in the Bilate catchment is shown in Figure 3. Considerable spatial variation mainly related to elevation differences in the watershed can be observed. The annual rainfall was higher (1162 mm to 1407 mm) in the highland areas north and west of the watershed than those in the lowlands, which received below 916 mm ( Figure 3a). The Kiremt rainfall declined from the northwest about 650 mm to the southeast of 320 mm (Figure 3b). In the Belg, the northern, central, and western parts received more than 400 mm (Figure 3c). The eastern and southeastern watersheds received less than 373 mm. The lowest seasonal rainfall throughout the watershed, below 230 mm, Climate 2021, 9, 98 8 of 22 was recorded during the Bega (Figure 3d). This spatial variation in rainfall distribution in the watershed is mainly attributed to the orographic effect.
The spatial distribution of annual and seasonal precipitation in the Bilate catchment is shown in Figure 3. Considerable spatial variation mainly related to elevation differences in the watershed can be observed. The annual rainfall was higher (1162 mm to 1407 mm) in the highland areas north and west of the watershed than those in the lowlands, which received below 916 mm ( Figure 3a). The Kiremt rainfall declined from the northwest about 650 mm to the southeast of 320 mm (Figure 3b). In the Belg, the northern, central, and western parts received more than 400 mm ( Figure 3c). The eastern and southeastern watersheds received less than 373 mm. The lowest seasonal rainfall throughout the watershed, below 230 mm, was recorded during the Bega (Figure 3d). This spatial variation in rainfall distribution in the watershed is mainly attributed to the orographic effect.

Temperature Variability Analysis
Temperature plays a vital role in water loss over a catchment. Increasing temperature reduces soil moisture by evapotranspiration and subsequently affects runoff amount due to drier soils [18]. Annual and seasonal temperature were analyzed from 1986 to 2015 to assess temperature variability in the watershed. The daily minimum temperature varied from the lowest of 10.9 °C in the highland area around Hossana to the highest of 30.38 °C in the lowlands of the watershed nearby Bilate Tena.
The CV of annual minimum temperature was 6%, which is much higher than those of average temperature (2.7%) and the maximum temperature (1.5%). A high variability in the minimum temperature over the past 30 years was observed in the study area. Figure  4 depicts anomalies of annual maximum, average, and minimum temperatures. The slopes of linear-fitting regressions for the minimum and average temperatures were higher than those for maximum temperature. A rise in the minimum temperature was more significant than that in the maximum temperature. In general, the increase in temperatures was consistent among the three temperatures, and the minimum temperature had a relatively persistent positive anomaly after 1994 than the others. Similar findings were observed in the Southern Ethiopia region adjacent to our study area [73].

Temperature Variability Analysis
Temperature plays a vital role in water loss over a catchment. Increasing temperature reduces soil moisture by evapotranspiration and subsequently affects runoff amount due to drier soils [18]. Annual and seasonal temperature were analyzed from 1986 to 2015 to assess temperature variability in the watershed. The daily minimum temperature varied from the lowest of 10.9 • C in the highland area around Hossana to the highest of 30.38 • C in the lowlands of the watershed nearby Bilate Tena.
The CV of annual minimum temperature was 6%, which is much higher than those of average temperature (2.7%) and the maximum temperature (1.5%). A high variability in the minimum temperature over the past 30 years was observed in the study area. Figure 4 depicts anomalies of annual maximum, average, and minimum temperatures. The slopes of linear-fitting regressions for the minimum and average temperatures were higher than those for maximum temperature. A rise in the minimum temperature was more significant than that in the maximum temperature. In general, the increase in temperatures was consistent among the three temperatures, and the minimum temperature had a relatively persistent positive anomaly after 1994 than the others. Similar findings were observed in the Southern Ethiopia region adjacent to our study area [73].

Rainfall Variability Analysis
Results of seasonal rainfall variability analysis showed higher CVs of 59% and 31% in the Bega and the Belg, respectively, than the 21% in the Kiremt. Regarding spatial variations, higher annual and seasonal rainfall CVs were mainly observed in lowlands around the Bilate and Bilate Tena stations having relatively lower rainfall amounts. The highland areas around Hossana and Fonko stations exhibited less variability than the lowlands. These results agree with studies in the Rift Valley Lake Basin [33,74] and across Ethiopia [37].

Rainfall Variability Analysis
Results of seasonal rainfall variability analysis showed higher CVs of 59% and 31% in the Bega and the Belg, respectively, than the 21% in the Kiremt. Regarding spatial variations, higher annual and seasonal rainfall CVs were mainly observed in lowlands around the Bilate and Bilate Tena stations having relatively lower rainfall amounts. The highland areas around Hossana and Fonko stations exhibited less variability than the lowlands. These results agree with studies in the Rift Valley Lake Basin [33,74] and across Ethiopia [37].
In addition to the CVs, the SAI was employed to assess annual rainfall variability in time domain. The annual and seasonal rainfall anomalies are presented in Figure 5. The most considerable annual positive rainfall deviation of 1.49 SAI was observed in 1997, while the largest negative anomaly of −2.91 SAI was detected in 2015 ( Figure 5a). The return period of wet/dry years was roughly two to three years as positive/negative anomalies appeared in Figure 5 for annual and seasonal rainfall, which is similar to the results presented in [30]. Since the 1990s, it seems that the appearance of negative anomalies are more distinct, showing a trend of having much less rainfall in dry years (Figure 5a). Similar findings were exhibited in the Rift Valley Lakes Basin [33] and Ethiopia overall [75].
As presented in Figure 5, the seasonal rainfall SAI revealed that 43.3% of the study period exhibited below-average rainfall in the Kiremt season. The Belg and the Bega showed that 53.3% of the study period rainfall was below the long-term average. Our findings also demonstrated that the dry Kiremt season is associated with El Niño events, and the wet season is associated with La Niña events; the opposite is true for the Belg.  In addition to the CVs, the SAI was employed to assess annual rainfall variability in time domain. The annual and seasonal rainfall anomalies are presented in Figure 5. The most considerable annual positive rainfall deviation of 1.49 SAI was observed in 1997, while the largest negative anomaly of −2.91 SAI was detected in 2015 ( Figure 5a). The return period of wet/dry years was roughly two to three years as positive/negative anomalies appeared in Figure 5 for annual and seasonal rainfall, which is similar to the results presented in [30]. Since the 1990s, it seems that the appearance of negative anomalies are more distinct, showing a trend of having much less rainfall in dry years (Figure 5a). Similar findings were exhibited in the Rift Valley Lakes Basin [33] and Ethiopia overall [75].

Streamflow Variability Analysis
Patterns of seasonal streamflow variability were similar to those observed in rainfall. The streamflow CVs were 51% and 47% in the Belg season and Bega, season, respectively, which were higher than those of annual (35%) and Kiremt season (32%). A comparable result was reported in the Rift Valley region of Ethiopia [19]. The higher variability of As presented in Figure 5, the seasonal rainfall SAI revealed that 43.3% of the study period exhibited below-average rainfall in the Kiremt season. The Belg and the Bega showed that 53.3% of the study period rainfall was below the long-term average. Our findings also demonstrated that the dry Kiremt season is associated with El Niño events, and the wet season is associated with La Niña events; the opposite is true for the Belg.

Streamflow Variability Analysis
Patterns of seasonal streamflow variability were similar to those observed in rainfall. The streamflow CVs were 51% and 47% in the Belg season and Bega, season, respectively, which were higher than those of annual (35%) and Kiremt season (32%). A comparable result was reported in the Rift Valley region of Ethiopia [19]. The higher variability of seasonal streamflow, especially in the Belg, which is one of the crop growing seasons, exacerbates the scarcity of available water and challenges in agricultural activities.
Interannual streamflow variability was the most critical characteristic in the Bilate watershed. Figure 6 shows the streamflow SAI of Bilate gauging station from 1986 to 2015. The number of years having negative anomalies was more than those with positive SAI, suggesting that the watershed was experiencing more dry years than wet ones. The findings also designated the 2000s as drier years than the 1990s. Taking years after 2000 as an example, there were 12 years with negative streamflow SAIs, which were more than nine years with negative rainfall SAIs (Figure 5a). The Bilate catchment is getting drier in streamflow.

Streamflow Variability Analysis
Patterns of seasonal streamflow variability were similar to those observed in rainfall. The streamflow CVs were 51% and 47% in the Belg season and Bega, season, respectively, which were higher than those of annual (35%) and Kiremt season (32%). A comparable result was reported in the Rift Valley region of Ethiopia [19]. The higher variability of seasonal streamflow, especially in the Belg, which is one of the crop growing seasons, exacerbates the scarcity of available water and challenges in agricultural activities.
Interannual streamflow variability was the most critical characteristic in the Bilate watershed. Figure 6 shows the streamflow SAI of Bilate gauging station from 1986 to 2015. The number of years having negative anomalies was more than those with positive SAI, suggesting that the watershed was experiencing more dry years than wet ones. The findings also designated the 2000s as drier years than the 1990s. Taking years after 2000 as an example, there were 12 years with negative streamflow SAIs, which were more than nine years with negative rainfall SAIs (Figure 5a). The Bilate catchment is getting drier in streamflow.

Changing Point Detection Tests
The Pettitt's, SNH, and BR test were adopted to detect shifts of hydroclimatic variables in annual and seasonal temperature, rainfall, and streamflow. Abrupt changes can be attributed to changing climate and human activities such as large water development projects [32]. In this study, changing points in temperature, rainfall, and streamflow in the Bilate watershed were examined for the period 1986-2015.

Temperature Changing Point Detection Test
As shown in Figure 7, temperature shifts were detected in 1993 for both annual and Belg season by three different tests. The annual mean temperature before and after the changing point were 18.59 • C and 19.50 • C, respectively. During the Kiremt season, mean temperatures of 17.79 • C and 18.68 • C were observed before and after, respectively, the changing point of 1994. After the abrupt change, the Belg and Bega temperatures increased by 0.98 • C and 0.85 • C, respectively. Such considerable temperature changes might be caused by droughts in Ethiopia observed over the past three decades [75].

Changing Point Detection Tests
The Pettitt's, SNH, and BR test were adopted to detect shifts of hydroclimatic variables in annual and seasonal temperature, rainfall, and streamflow. Abrupt changes can be attributed to changing climate and human activities such as large water development projects [32]. In this study, changing points in temperature, rainfall, and streamflow in the Bilate watershed were examined for the period 1986-2015.

Temperature Changing Point Detection Test
As shown in Figure 7, temperature shifts were detected in 1993 for both annual and Belg season by three different tests. The annual mean temperature before and after the changing point were 18.59 °C and 19.50 °C, respectively. During the Kiremt season, mean temperatures of 17.79 °C and 18.68 °C were observed before and after, respectively, the changing point of 1994. After the abrupt change, the Belg and Bega temperatures increased by 0.98 °C and 0.85 °C, respectively. Such considerable temperature changes might be caused by droughts in Ethiopia observed over the past three decades [75].

Rainfall Changing Point Detection Test
Unlike the increase shift in mean temperature, abrupt changes in annual and Belg rainfall were detected in 1997 by all three tests as shown in Figure 8, while there was no significant change in Kiremt and Bega rainfall. Reduction in annual rainfall from 1246 mm to 1110 mm was mainly attributed to the decrease in the Belg rainfall from 502.3 mm to 378.0 mm. In the Bilate catchment, the Belg season is one of the rainy seasons, and accounts for about 36.5% of the total annual rainfall. The downward shift in rainfall during the Belg season since 1997 has resulted in extreme drought and increased mortality rate of cattle by 49% in southern Ethiopia around the Bilate watershed [78]. Because rainfall has a great impact on streamflow, the decline in rainfall can directly affect the streamflow and exacerbate water stress.

Rainfall Changing Point Detection Test
Unlike the increase shift in mean temperature, abrupt changes in annual and Belg rainfall were detected in 1997 by all three tests as shown in Figure 8, while there was no significant change in Kiremt and Bega rainfall. Reduction in annual rainfall from 1246 mm to 1110 mm was mainly attributed to the decrease in the Belg rainfall from 502.3 mm to 378.0 mm. In the Bilate catchment, the Belg season is one of the rainy seasons, and accounts for about 36.5% of the total annual rainfall. The downward shift in rainfall during the Belg season since 1997 has resulted in extreme drought and increased mortality rate of cattle by 49% in southern Ethiopia around the Bilate watershed [78]. Because rainfall has a great impact on streamflow, the decline in rainfall can directly affect the streamflow and exacerbate water stress.

Streamflow Changing Point Detection Test
Abrupt changes of reduction in streamflow were detected for annual and the Belg season in 2007 as shown in Figure 9.      The stationarity of rainfall and streamflow before and after the changing point was examined by frequency analysis, as shown in Figure 10 and Table 2. The magnitude of precipitation and streamflow of a particular exceedance probability decreased after an abrupt shifting point. For example, the rainfall (Figure 10a) reduced after the changing point by 6.7% and 18.2% for the exceeded probability of 0.1% and 50%, respectively. Frequency analysis of streamflow (Figure 10b) also attests that the magnitude after the changing point was lower than before the changing point. As streamflow equaled or exceeded 0.1% and 50% of the time, its magnitude dropped by 29.7% and 33.3%, respectively. This reduction showed that there is a shortage of available water in the Bilate watershed after the changing point. Therefore, appropriate water conservation structures are recommended to minimize the water scarcity issue. by 6.7% and 18.2% for the exceeded probability of 0.1% and 50%, respectively. Frequency analysis of streamflow (Figure 10b) also attests that the magnitude after the changing point was lower than before the changing point. As streamflow equaled or exceeded 0.1% and 50% of the time, its magnitude dropped by 29.7% and 33.3%, respectively. This reduction showed that there is a shortage of available water in the Bilate watershed after the changing point. Therefore, appropriate water conservation structures are recommended to minimize the water scarcity issue.

Implication of Changes in Hydrological Cycles
After detecting abrupt changes in temperature, rainfall, and streamflow, we further investigated the effects of such changes in evapotranspiration (ET) and basin water storage (ΔS) to understand the implication of changes in hydrological cycles in the Bilate watershed. Based on the years of shifts in hydroclimatic parameters, we divided the study years into four periods.
As presented in Figure 11, the findings indicated a large ET and small ΔS during 1986-1993. This low amount of ΔS was caused by a higher ET loss due to higher temperature differences between the maximum and minimum temperature. During 1993-1997, the ΔS exhibited an increasing pattern due to a reduction in ET. To examine which factor

Implication of Changes in Hydrological Cycles
After detecting abrupt changes in temperature, rainfall, and streamflow, we further investigated the effects of such changes in evapotranspiration (ET) and basin water storage (∆S) to understand the implication of changes in hydrological cycles in the Bilate watershed. Based on the years of shifts in hydroclimatic parameters, we divided the study years into four periods.
As presented in Figure 11, the findings indicated a large ET and small ∆S during 1986-1993. This low amount of ∆S was caused by a higher ET loss due to higher temperature differences between the maximum and minimum temperature. During 1993-1997, the ∆S exhibited an increasing pattern due to a reduction in ET. To examine which factor influences ET reduction with the increasing temperature in the watershed, we performed trend tests for ET, PET, and temperature terms in the Hargreaves equation. As shown in Table 3, the M-K test indicated a decreasing trend of PET and ET. The term (T max − T min ) in the Hargreaves equation showed a significant decrease by 0.046 • C/year while (T max + T min )/2 increased by 0.044 • C/year. The more significant decreasing rate in (T max − T min ) than the increase of (T max + T min )/2 resulted in reducing PET calculated by the Hargreaves method. This reduction in the PET leads to a decline in ET estimated by the Budyko-type relation. Reduction in rainfall after 1997 further induced a decrease in ET and ∆S due to limited water supply. From 1997-2015, a decline in rainfall further induced a decrease in ET, which outweighed the increasing temperature. Our results suggest that the scarcity of available water is getting worse, along with abrupt changes in temperature, rainfall, and streamflow. The accelerated reduction of ET and ∆S after 1997 may be related to severe drought caused by intensive El Niño events [80,81]. During 2007-2015, the ∆S exhibited an increasing pattern due to declined discharge as the ∆S is calculated as the residual of rainfall, streamflow, and evapotranspiration. the increase of (Tmax + Tmin)/2 resulted in reducing PET calculated by the Hargreaves method. This reduction in the PET leads to a decline in ET estimated by the Budyko-type relation. Reduction in rainfall after 1997 further induced a decrease in ET and ΔS due to limited water supply. From 1997-2015, a decline in rainfall further induced a decrease in ET, which outweighed the increasing temperature. Our results suggest that the scarcity of available water is getting worse, along with abrupt changes in temperature, rainfall, and streamflow. The accelerated reduction of ET and ΔS after 1997 may be related to severe drought caused by intensive El Niño events [80,81]. During 2007-2015, the ΔS exhibited an increasing pattern due to declined discharge as the ΔS is calculated as the residual of rainfall, streamflow, and evapotranspiration. Figure 11. Implication of changes in hydrological cycles of the Bilate watershed. The symbol ↑represents increasing trends while ↓ represents decreasing trends.

Temperature Trend Analysis
Results of the trend analysis indicate that the annual and seasonal maximum, average, and minimum temperatures increased significantly, as shown in Table 4. The annual minimum temperature rose faster (0.06 °C/year) than the maximum temperature (0.02 °C/year). Higher minimum temperature is beneficial to crops due to reduced chilling effects [82]. Previous studies in northern Ethiopia also found an increase in average maximum temperature of 0.04 to 0.051 °C/year from 1983 to 2016 [83]. Figure 11. Implication of changes in hydrological cycles of the Bilate watershed. The symbol ↑ represents increasing trends while ↓ represents decreasing trends. Table 3. M-K trend test of annual PET, ET, (T max − T min ), and (T max + T min )/2.

Temperature Trend Analysis
Results of the trend analysis indicate that the annual and seasonal maximum, average, and minimum temperatures increased significantly, as shown in Table 4. The annual minimum temperature rose faster (0.06 • C/year) than the maximum temperature (0.02 • C/year). Higher minimum temperature is beneficial to crops due to reduced chilling effects [82]. Previous studies in northern Ethiopia also found an increase in average maximum temperature of 0.04 to 0.051 • C/year from 1983 to 2016 [83].  Table 5 presents the M-K test and Sen's slope results of the seasonal and annual mean temperatures before the changing point (1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993), after the changing point , and full period . Significant increasing trends were detected for seasonal and annual mean temperature during the full period with the increasing rate of the Bega > the annual > the Belg > the Kiremt. Before the changing point, the seasonal mean temperature showed an insignificant increasing trend during the Bega and Kiremt seasons. After the changing point, all seasonal and annual temperatures exhibited a significant increasing trend, except for the Belg, which increased insignificantly with 0.01 • C/year. The Bega had the largest increasing rate before and after the changing point and during the full period. A study in Ethiopia between 1983 and 2012 [75] found a substantial increase in drought occurrences and temperature intensification. It is recommended that farmers should look for additional small-scale water sources and adopt drought-resistant crops to cope with the drought. Table 5. The M-K test and Sen's slope results of seasonal and annual mean temperatures before the changing point (1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993), after the changing point , and full study period .

Streamflow Trend Analysis
Seasonal and annual streamflow trends before the changing point , after the changing point (2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015), and during the entire study period  were examined and summarized, as shown in Table 7. The Kiremt season designated a significant reduction with 2.33 mm/year before the changing point and 7.15 mm/year after the changing point. Before the changing point, the Belg season and annual streamflow significantly decreased. The entire study period exhibited a statistically significant decreasing trend for the annual streamflow with a 3.64 mm/year. The seasonal time scale also significantly decreased, with the most considerable reduction of 2.04 mm/year in the Kiremt season. This finding indicates that the amount of surface water in the watershed decreases from time to time. The decline in streamflow in the watershed is mainly caused by reduced rainfall. Such a significant reduction of available water resources can result in water stress and failure of crop yield. Table 7. The M-K test and Sen's slope results of seasonal and annual mean streamflow before the changing point , after the changing point (2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015), and full study period .

Seasons
Full

Variability Analysis of Hydroclimatic Variables
With enormous domestic and agricultural water demands [26,28], the Bilate watershed is highly water-stressed and vulnerable to climate change. The Belg (FMAM) and the Kiremt (JJAS) seasons account for 85% of annual rainfall, with peaks in July and August and the lowest was in December (Figure 2). With the mean minimum of 11.25 • C to the mean maximum of 27.86 • C, the study area has a high diurnal variation. The high diurnal temperature variation of Ethiopian river basins was also found in the Gilgel Abay watershed of the Blue Nile Basin [14]. Temperature anomaly also designated the increased frequency of hot days and nights than cold days and nights. Similar findings were exhibited in the Blue Nile Basin [50] and the entirety of Ethiopia [8,75].
In the Bilate watershed, the El Niño events are negatively correlated with the Kiremt rainfall and positively with the Belg rainfall. However, the La Niña episodes showed a positive relationship with the Kiremt and a negative with the Belg rainfall. As shown in the previous studies, the weakening of Kiremt rain resulted in a devastating drought and the 1984 famine in Ethiopia was correlated with El Niño [84]. The rainfall anomaly also witnessed interannual variability in the Bilate watershed with more profoundly belowmean rainfall since the 1990s, which indicated that the study area was relatively drier in the 20th century. Recent studies in the Rift Valley Lakes Basin, which encompasses the Bilate watershed, also showed higher rainfall variability and frequent drought events [29,33]. The rainfall variability and reduction in amount affect the agricultural activities over the Bilate watershed [30]. Annual and seasonal streamflow variability is consistent with the rainfall, which is in line with the results presented in [76,84,85]. Therefore, factors that modulate the rainfall variability should be considered when studying the inter-annual streamflow alterations in the Bilate.

Detection of Hydroclimatic Variables Changing Points
The abrupt change in hydroclimatic variables showed an upward shift in temperature and a downward change in rainfall and streamflow. The changing point of annual and seasonal mean temperature occurred in 1993. Similarly, the Kulfo watershed in the southern part of the Bilate watershed exhibited an average temperature increasing shift in 1993 [32]. This abrupt increase in temperature in the area might contribute to droughts in the 1990s. Ethiopia suffered a drought in 1993/94, and food shortages reached five million tons [86]. We found similar results in studies in the nearby Gidabo catchment, Rift Valley Lakes Basin [18], and the Omo-Ghibe Basin west of the Rift Valley Lakes Basin [31].
Unlike temperature, rainfall and streamflow showed a sudden downward shift in the Bilate watershed. Belg and annual rainfall designated a significant decreasing change in 1997 by 24.7% and 11%, respectively. The decrease in rainfall after 1997 in the Bilate resulted in lower food production and disease outbreaks such as malaria and cholera [36].
Studies in the nearby Gidabo catchment [18], Omo-Gibe Basin [31], and the Upper Blue Nile Basin [22] presented comparable findings. The streamflow analysis identified a downward shift with a more considerable reduction in the Belg season by 53% after the changing point. This downward shift also resulted in one of the most extreme drought years in history, mainly affecting southern Ethiopia. About 26,000 livestock died during the 2008 drought in the region [87]. The decrease in streamflow was more significant than rainfall. The higher streamflow reduction than rainfall might result from a combined effect of a decline in the rain and other anthropogenic impacts [82,88].
The changing points in the study area could be related to climatic factors such as El Niño and La Nina events. For instance, the decreasing shift in rainfall after 1997 in the Bilate watershed matched with the 1997/98 intense El Niño year [36]. As La Niña reduces the Belg season rainfall in southern Ethiopia, the abrupt downward shift of the Bilate watershed streamflow in 2007/2008 attributed to the intense La Niña episodes [89]. Therefore, we suggest that the shifting points captured by the changing point detection methods in the Bilate watershed are related to these ENSO events. Similarly, the ENSO phenomenon shifted climatic variable patterns over different regions of the world. For example, studies on surface air temperatures in Saudi Arabia [90] and climate extremes in southern China [91] showed a significant change for most stations in the 1997/1998 El Niño year. In addition, studies in Malaysia exhibited a changing point of rainfall and temperature in 1997/98 and in 2001/2002, which were related to El Niño and La Niña events [92,93]. However, the other oscillations like Atlantic Multi-decadal Oscillation (AMO) and Indian Ocean Dipole (IOD) have a weak correlation with southern Ethiopian rainy seasons [37,84]. In addition, [94] revealed that AMO and IOD had less influence on the urban center of Ethiopian rainy seasons and annual rainfall extremes. Therefore, the shifting points captured by the changing point detection methods are less likely related to AMO and IOD.

Hydroclimatic Variables Trend Analysis
To examine the effects of changing points on the trend of hydro-climatic variables in the Bilate watershed, we carried out an M-K test for the entire period, before and after the changing point. The M-K test showed a statically insignificant trend except in the Belg season for temperature before the changing point (Table 5) and for the streamflow after the changing point except in the Kiremt season (Table 7), but it exhibited a significant trend in all seasonal and annual basis for the entire study period. This insignificant trend in most seasons before and after the changing point indicates a decrease in the statistical significance of hydroclimatic variable trends as the data years are limited or below ten years, which the World Meteorological Organization (WMO) recommends for data-scarce regions [95]. Correspondingly, a study in Chile estimating annual maximum and minimum flow trends [96] revealed higher uncertainty and less significance of trend results for the limited data length than longer years.
The Rift Valley Lakes Basin encompasses Bilate, Kulfo, and other watersheds with varied topography and climate patterns. Recent hydroclimatic variable trend studies on the two watersheds, the Kulfo [32] in the south and the Mojo [34] north of the Bilate watershed, were compared with our findings. The Bilate, Kulfo, and Mojo watersheds drainage areas are 5518 km 2 , 127.09 km 2 , and 1984.96 km 2 . The following discussions are based on numbers presented in Table 8 adopted from data presented in recent studies for the Kulfo [32] and the Mojo [34] watersheds.
The increasing trend of annual average temperature in the Bilate watershed is higher than that in the Kulfo (Table 8). The increase in temperature overall Ethiopia is 0.025 • C/year since 1960 [8]. This study's trend test revealed that the climate has become warmer in the Bilate watershed during the past 30 years than in the other catchments. The annual mean minimum temperature is more profoundly increasing in the Bilate than the other two watersheds. The increasing rate in the minimum temperature in the Bilate watershed is higher than the maximum, which agrees with the findings in central Ethiopia [8]. The higher increasing rate in the minimum temperature indicated that the overall increase in the watershed's annual temperature is mainly attributed to the minimum temperature increase. The Bilate had a decrease in annual rainfall by 8.32 mm/year, but the Mojo and the Kulfo were increased by 3.85 mm/year and 3.89 mm/year, respectively. The most considerable decrease in the Belg rainfall in the Bilate accounted for a declining trend in the annual rainfall. However, the increase in annual rainfall in the Mojo watershed was attributed to the increase in Kiremt rainfall, and the Kulfo is regulated by the Bega rainfall ( Table 8). The substantial decrease in annual and seasonal rainfall in the Bilate watershed indicated that the watershed is drier than the Mojo and the Kulfo watersheds.
The Bilate and the Mojo watersheds' annual mean streamflow decreased by 3.64 mm/year and 0.04 mm/year. However, the Kulfo showed an increasing trend by 0.25 mm/year. During the Kiremt season, the streamflow reduction in the Bilate is more than that in the Mojo. On the other hand, the Kulfo showed a slightly increasing trend (Table 8). A declining and increasing annual streamflow trend in the Bilate and Kulfo watershed was consistent with the annual rainfall trend. However, there was a slight reduction in annual streamflow as rainfall increased in the Mojo watershed due to overexploitation of water for industrial activities [79]. The more considerable streamflow decrease in the Bilate watershed than the other two watersheds implies that the Bilate is profoundly under water scarcity than the Mojo and Kulfo watersheds. Therefore, it is highly recommended to adopt watershed management schemes against water scarcity issues in the Bilate watershed.

Conclusions
This study investigated the change and variability of rainfall, temperature, and streamflow in the Bilate watershed from 1986 to 2015. The data were analyzed using the coefficient of variation, standardized anomaly index, Mann-Kendall trend test, Sen's slope estimator, and three changing point detection tests (Pettitt, SNH, and BR). Variations in rainfall and streamflow are consistent of having high variability during the Belg and the Bega seasons compared to the Kiremt season. This higher seasonal rainfall and streamflow variation, especially in the Belg season, one of the crop growing seasons, impacts water availability and agricultural production. The variability of the maximum and the minimum temperatures were less significant than those of the rainfall and streamflow. However, the basin has experienced drier conditions since the 1990s, which indicates that water scarcity in the 20th century in the Bilate watershed was more severe than it was in previous years.
Results of changing point analysis in rainfall and streamflow exhibited a significant decreasing change in the annual and Belg time series at a 5% significance level. However, temperature designated an upward shift. The changing point analysis revealed that the rainfall and streamflow detection points of both annual and Belg season shifting points occurred in the 1997/1998 and 2007/2008 El Niño and La Niña years. These changing points are possibly related to ENSO episodes.
The Mann-Kendall trend test and Sen's slope estimator showed a significant increasing temperature trend with a faster increasing rate of minimum temperature than the maximum. This indicates the overall rise in the watershed's annual temperature, mainly attributed to the minimum temperature increase. On the other hand, the annual and Belg rainfall exhibited a substantial decrease in the entire study period, while the Bega and Kiremt showed an insignificant decreasing trend. Besides, the annual and seasonal streamflow revealed a significant decreasing trend. Thus, a consistent decreasing pattern in annual and seasonal rainfall and streamflow was observed. The significant increase in temperature and substantial decline in rainfall caused streamflow reduction and resulted in water scarcity in the Bilate watershed.
The Bilate River is the primary water source for domestic and agricultural usages of the surrounding communities. Thus, the erratic nature of rainfall and water scarcity could severely affect agricultural productivity during the crop growing seasons. Therefore, concerned bodies in the Bilate watershed should construct appropriate multipurpose infrastructures such as reservoirs and water harvesting ponds to store the excess water in the rainy seasons for the dry season. The reduction and variability of streamflow in the Bilate catchment require urgent considerations of adopting better water management strategies, developing adaption measures against impacts of climatic change, and implementing regulations for watershed conservations. Results of this study improved our understanding of annual and seasonal hydroclimatic variability in the Bilate watershed. Further investigation is recommended to quantify land-use and climate change effects using rainfall-runoff modeling and proper utilization of the scarce water resources in the watershed.