Detection and modeling of streamflow variations in the Abay (upper Blue Nile) basin, Ethiopia Open Access

The discharge in rivers, streams, and other natural channels is measured as streamflow. It is also essential to the hydrosphere because streams transport mass and energy through watersheds. Furthermore, it is an important parameter for studying drought, floods, and the management of water resources (Modarres 2007; Myronidis et al. 2018). According to Gebremicael et al. (2019), the hydrology of basins and the variability of this hydrology can be comprehensively understood using appropriate statistical tools to investigate long-term streamflow patterns.

Plans and decisions regarding basin management cannot rely solely on statistical methods. Therefore, it is essential to model the changes hydrologically before making decisions about basin management and action plans. Streamflow is essential to the hydrological cycle, which keeps natural water systems' water mass balance (Rusjan & Mikoš 2015; Heerspink et al. 2020). Compared with point data, like precipitation data, streamflow is more appropriate for discovering regional patterns for river basins since it reflects spatial information about a watershed (Yu et al. 2011; Nalley et al. 2012; Henn et al. 2018).

For the construction, development, and operation of various kinds of water facilities, streamflow data are crucial (Hughes 2001; Chiew et al. 2003). An annual streamflow series is deemed steady for those purposes if its mean remains constant and its autocorrelation is time-independent (Tian et al. 2011; Tarpanelli et al. 2012). However, under changes in climate, structural design, and land use and land cover (LULC), these requirements might not be achieved. It is important to provide plausible explanations for trends in hydrological indicators that have been noticed, such as sudden changes. Managing water resources globally requires an understanding of trends in hydrological variables and the measurement of streamflow fluctuation patterns (Sanborn & Bledsoe 2006; Liu et al. 2017).

It is necessary to gather data and model streamflow variance to support effective water resource management. Managers of water resources are starting to create sustainable water management systems that can adapt to changes in streamflow. Ethiopia's water resources are highly vulnerable to climatic and hydrological variability because of variable topography, high population density, land degradation, and inadequate implementation of water management techniques (Bogale & Tolossa 2021). The Abay (upper Blue Nile basin) in Ethiopia is one of the major tributaries of the Nile River, accounting for more than 60% of the river's total discharge (Abate et al. 2015; Abiy et al. 2016; Mengistu et al. 2021).

Many researchers have found that LULC changes considerably influence the streamflow regime of the Abay basin, causing an increase in the variability of river flows (Adem et al. 2016; Mengistu et al. 2021; Malede et al. 2022). Tekleab & Kassew (2019) showed that changes in hydrological regimes brought about by LULC alterations had an impact on peak flows, streamflow patterns, and flow volume. Soil erosion brought on by variations in streamflow reduces the fertility of the soil in upper catchments and also causes sedimentation in irrigation canals and reservoirs downstream (Nepal et al. 2014). The majority of the streamflow variability in the previous basins is examined using a single station data, a single flow series, and a single analysis (either statistical or modeling). However, using a single station data, a single flow series, and a single analysis (either statistical or modeling) to investigate the streamflow variability across the entire basin is challenging.

All of the aforementioned characteristics are present in the streamflow of the Abay basin; however, the streamflow variation in this basin has not been thoroughly examined by gathering data from numerous stations, multiple flow series, hydrological modeling, and statistical streamflow analysis. Studying the streamflow changes by gathering data from numerous stations, multiple flow series, statistical streamflow analysis, and hydrological modeling is important to improve the knowledge and understanding of the Abay basin's streamflow variations.

To make an immediate water management choice, the current study modeled significantly changed stations in the Abay basin and identified stations with considerable streamflow variations among 29 selected stations in three flow (mean annual, annual maximum, and 7-day average minimum) series. This is important because the basin plays a significant role in the Nile River's flow, and researchers and planners can use it as a guide to develop suitable plans for addressing the basin's increasing water demand. Modeling streamflow variations across multiple stations and flow series contributes insights for water managers and decision-makers in developing sustainable water resource management strategies for addressing the basin's increasing water demand. Additionally, it will provide information for future researchers and planners in the basin to examine changes in streamflow.

A semi-distributed hydrological model, the Hydrologiska Byråns Vattenbalansavdelning (HBV)-Light model, has been developed for water balance studies and hydrological modeling. In the 1970s, the Swedish Meteorological and Hydrological Institute's Water Balance Division developed this model initially to support hydropower operations. In 1993, Uppsala University modified the HBV-Light model by using Microsoft Visual Basic (Seibert & Vis 2012; Ali et al. 2018); this modified model applies to more than 50 countries worldwide.

Several scientists have used the HBV-Light model for hydrological applications in the Abay (upper Blue Nile) basin, and their results indicate that this model is of satisfactory effectiveness. Rientjes et al. (2011) employed the HBV-Light model to simulate the water level in the Lake Tana sub-basin in Ethiopia and found that the Nash–Sutcliffe efficiency (NSE) during the analysis period was as high as 0.91; thus, the model exhibited good performance. Worqlul et al. (2015) compared the performance of the HBV and parameter-efficient distributed (PED) models when applied to the Gilgel Abay catchment of the Abay basin.

They concluded that the HBV-Light model performed somewhat better than the PED model because it increased the number of calibration parameters by dividing the watershed into sub-basins. Abebe & Kebede (2017) assessed the effect of climate change on the water supply of the Megech watershed in Ethiopia by using the HBV-Light model; they reported that this model had satisfactory overall performance in the calibration and validation phases (NSE = 0.91). The HBV-Light model was also utilized for this study to make sure that the basin's topography characteristics such as slope, elevation, and vegetation distribution were taken into account.

The Abay basin comprises agropastoral land, agricultural land, marshland, urban land, cultivated land, forest land, and grassland with frequent clusters of shrubs, woodlands, trees, and water. The Abay basin's climate is semi-arid to arid and is mostly impacted by the seasonal migration of the intertropical convergence zone (ITCZ). Rainfall in the basin is very seasonal. This seasonality influences a wide range of local climates, from hot and arid near the Ethiopia–Sudan border to temperate in the highlands and even humid-cold at Ethiopia's mountain peaks. The annual rainfall in this basin varies, with the Didessa and Dabus sub-basins receiving 2,200 mm and the Ethiopia–Sudan border receiving 900 mm (Yasir et al. 2014).

Precambrian foundation rock and volcanic rock dominate the geology of the Abay basin, with some sedimentary rock also present (Mengistu et al. 2021). The Abay basin's flow is primarily seasonal, which reflects the seasonal rainfall in the basin at different stations. Moreover, the streamflow measurement stations in this basin cover areas of various sizes, with the smallest and largest areas covered by the Komise (112 km²) and Kessie (64,882 km²) stations, respectively (Supplementary Table S1). The Kessie station covers an area located between the latitudes 9°12′18″ and 12°45′20″ N and between the longitudes 36°43′55″ and 39°49′12″ E.

Data on daily streamflow for the period 6:00 a.m. and 6:00 p.m. at 36 stations in the Abay basin were obtained from the Abay Basin Authority (Supplementary Table S1). The Abay Basin Authority also provided the land cover map for the Kessie watershed. Precipitation, temperature, humidity, and sunshine hour data were acquired from the Hydrology Department of the Ethiopian Ministry of Water, Irrigation, and Energy.

Following careful data screening and quality checks, 29 stations were selected for hydrological analysis based on their long-term data availability (Supplementary Table S2). Table 2 lists the number of stations with different periods of data availability after the quality inspection was conducted.

As required by the semi-distributed HBV-Light model, a digital elevation map of the Kessie Watershed was created using Shuttle Radar Topography Mission data with a resolution of 30 m × 30 m. The sub-basin area was then classified into various hydrological units based on elevation ranges (i.e., elevation zones). This step was conducted to ensure that the topographic parameters of the study area including slope, elevation, and vegetation distribution were considered in the HBV-Light model.

The HBV-Light model requires daily rainfall data. Consequently, 12-year rainfall data (2002–2013) for 11 meteorological stations in the Kessie watershed were collected, and these stations were selected according to their weighted contribution to the Kessie watershed (Table 3). The Arc-GIS software program was used to multiply the rainfall by the weight obtained from a Thiessen polygon to determine the areal rainfall.

Because the HBV-Light model only permits the creation of three vegetation zones in a catchment area, the five land cover groups in the land cover map of the Kessie watershed were divided into three groups (Hundecha & Bárdossy 2004). Finally, to obtain catchment settings for the HBV-Light model, the vegetation zones were overlaid on the elevation zone map by using zonal statistics.

Hydrological models can be manually calibrated through a process of trial and error by the method of Bergström (1992). Sufficient data that cover a range of hydrological events during the calibration period should be used in the aforementioned method. To evaluate the fit between the simulated and observed discharge, plots of the parameters were visually evaluated, the accumulated difference was computed, and statistical criteria were used.

In the Abay basin, the mean annual flow varied between 1.89 (at the Indris station) to 576.85 m³/s (at the Kessie station; Supplementary Table S3). Moreover, the standard deviation ranged from 0.92 (Indris) to 155.84 m³/s (Kessie). The coefficient of variation varied between 0.16 (Bahir Dar and Gilgel Abay) and 0.34 (Chemoga). In addition, the skewness of this flow varied from −0.78 (Guder) to 3.82 (Muger). Finally, the kurtosis of the mean annual flow ranged from −1.43 (Didessa) to 15.27 (Muger).

As presented in Supplementary Table S4, during the analysis period, the maximum annual flow varied between 3,696.58 (Kessie) and 13.60 m³/s (Hoha). Furthermore, the standard deviations ranged from 14.09 (Neshi) to 1,293.36 m³/s (Kessie). The coefficient of variation varied between 0.22 (Gumera) and 1.76 (Chemoga). In addition, the skewness of this flow varied from −0.48 (Sibilu) to 4.1 (Robigumero). Excessive kurtosis was discovered for the Komise and Bello stations kurtosis values of −1.69 and 20.12, respectively.

The minimum streamflow averaged over 7 days per year ranged from 0.01 (Robijida and Dondor) to 79.25 m³/s (Kessie; Supplementary Table S5). Moreover, the standard deviation ranged from 0.02 (Dondor, Robigumero, and Robijida) to 56.49 m³/s (Kessie). The coefficient of variation ranged from 0.2 (Andassa) to 3.86 (Muger). In addition, the skewness values ranged from −0.21 (Guder) to 4.73 (Ribb). Finally, the kurtosis values of this parameter varied from −0.91 (Uke) to 23.13 (Ribb).

The basin typically has high runoff during the rainy season, which runs from June to September, and lower runoff during the dry season (Supplementary Table S6). The highest mean monthly flow occurred in August and September at 25 (86.21%) and 4 (13.79%) respectively. The Kessie station had the highest mean monthly flow of all stations (2,329.3 m³/s in August). At all stations except Dondor, the mean monthly streamflow was lowest in February; at Dondor, it was lowest (0 m³/s) in April. The smallest standard deviation in the mean monthly streamflow was 0.0 m³/s, which was for the Robijida and Aletu stations in February and December, respectively. Moreover, the highest standard deviation was 628.02 m³/s, which was the value for the Kessie station in August.

The lowest coefficient of variation for the mean monthly flow was 0.2; this value was obtained for Gumera, Guder, Gilgel Abay, Dabana, and Sibilu in August; Gilgel Abay in September; and Dabana in July. The highest coefficient of variation was 3.7; this value was obtained for Robigumero in June. The highest and lowest skewness values for the mean monthly streamflow (5.7 and −1.2, respectively) were determined for the Bello station in March and August, respectively. Finally, the highest kurtosis value for the mean monthly streamflow (35.7) was observed for Guder in March, whereas the lowest kurtosis value (−1.7) was obtained for Fettam in June.

Several change points were identified among the 29 selected stations in the Abay basin in the mean annual flow. Notable changes in streamflow over time were discovered for 14 stations (48.28%). Six of these 14 stations (42.86%) Didessa, Dondor, Hoha, Megech, Temcha, and Uke (Supplementary Figure S1(a)) exhibited the most significant shifts (p-values of 0) throughout the analysis (Supplementary Table S10).

The variations in the annual maximum streamflow of 13 stations (44.83%) indicated significant shifts at various times. Of these stations, Kessie, Bello, Dondor, Gilgel Beles, and Temcha (Supplementary Figure S1(b)) had the most significant shifts (p-values of 0) throughout the analysis (Supplementary Table S11).

The variations in the annual 7-day average minimum streamflow of 10 stations (34.38%) revealed significant shifts at different times (Supplementary Table S12). The most significant shifts (p < 0.01) occurred for four stations (40%): Kessie, Chemoga, Temcha, and Uke (Supplementary Figure S1(c)).

The streamflow variation in the Kessie watershed was modeled using two LULC maps (2002 and 2013) to evaluate the performance of the HBV-Light model and to ascertain how LULC affects streamflow when all other factors are constant. In the calibration and validation phases, the HBV-Light model exhibited satisfactory performance for the aforementioned watershed.

Sensitivity analysis was conducted to understand the effects of different model parameters on the performance of the HBV-Light model. According to Rientjes et al. (2011), the NSE is most sensitive to the following parameters: FC, LP, and BETA. Temesgen (2019) stated that K2 and MAXBAS are the main factors influencing the performance of the HBV model. In the present study, sensitivity analysis was conducted using the KGE.

Each model parameter's effect on the performance of the HBV-Light model was graphically plotted and visualized; the parameters in the graphs exhibiting the fastest slope changes had the strongest influences on the model's performance. Based on the plotted results, the performance of the HBV-Light model was found to be highly sensitive to FC, PERC, Ko, and K2 but less sensitive to LP, K1, and BETA.

The model was calibrated and validated for 12 years (2002–2013), 1 year for warming-up period (1 January 2002–31 December 2002), 7 years for calibration (2003–2009), and 4 years for validation (2010–2013). To find the parameters that would best match the simulated discharges of the two LULC maps, the snow routine settings were reset to zero for several rounds. This step was taken to maximize the model's performance in simulating streamflow data under particular LULC conditions.

As it is analyzed by Huziy et al. (2013) streamflow characteristics to predict extreme weather events and seasonal streamflow in their study area increased or decreased by the season, whereas the mean annual streamflow increased significantly over time almost throughout the study area.

This study also analyzed the statistical characteristics of streamflow at different temporal scales for 29 selected stations in the Abay basin in Ethiopia to know the streamflow features and make informed decisions. As presented in Supplementary Tables S3 and S4, different characteristics on an annual scale were discovered for different stations in the Abay basin. The majority of stations have a discharge between 0 and 10 m³/s in mean annual streamflow and a maximum annual streamflow of greater than 100 m³/s at 17 stations of the 29 stations.

The annual 7-day average minimum streamflow was less than 1 m³/s for 23 stations (79.31%), which suggests that almost all tributaries of the Abay basin are seasonal. This is important to make informed decisions in the basin water. Moreover, the highest mean monthly flows occurred in August for 25 stations (Supplementary Table S6). Finally, as presented in Supplementary Tables S3–S6, Kessie had the highest maximum annual and monthly streamflow values of all stations considered in this study.

Lacking confirming the hydrological model with the statistical result, Tekleab et al. (2013) attempted to examine the streamflow trends and change points for nine stations in the Abay basin and found that the mean annual streamflow was trending both downward and upward. Mekonnen et al. (2018) also analyzed annual and seasonal streamflow variations, noting significant increasing trends in both annual and seasonal streamflow.

This study selected 29 stations in the Abay basin and identified significant variations in streamflow to conduct hydrological modeling. It also generated a three time series (annual mean, annual maximum, and 7-day average minimum flow) to understand its trends and change points effectively. For all 29 stations, an increasing or decreasing trend in mean annual streamflow was observed, and nearly half of these trends were significant. Most of the significant trends in mean annual streamflow were increasing trends (Supplementary Table S7).

As presented in Supplementary Table S8, for every station except for Muger (no change), an increasing or decreasing trend in annual maximum streamflow was observed. A total of 15 and 13 stations showed decreasing and increasing trends, respectively, with five stations each having significant decreasing and increasing trends. All stations except for Bahir Dar (no change) had an increasing or decreasing trend (Supplementary Table S9) in the annual 7-day average minimum streamflow flow. A total of 20 and eight stations had increasing and decreasing trends, respectively, with seven and four stations each having significant increasing and decreasing trends, respectively (Supplementary Table S9).

The result of this study pointed out that a significant variation in mean annual flow was discovered for 14 stations (Supplementary Table S10), with six stations equally having the largest changes (p-values of 0): Didessa, Dondor, Hoha, Megech, Temcha, and Uke (Supplementary Figure S1(a)). The significant changes in annual maximum streamflow in different periods were discovered for 13 stations (Supplementary Table S11), with five stations showing the largest variations (p-values of 0): Abay at Kessie, Bello, Dondor, Gilgel Beles, and Temcha (Supplementary Figure S1(b)). Tekleab et al. (2013) identified a significant decreasing trend in the annual 7-day average minimum streamflow in the Gilgel Abay catchment around 1994. However, this study found that Gilgel Abay exhibited an increasing trend in 1990. In addition to this, only 10 stations exhibited significant variation in annual 7-day average minimum streamflow (Supplementary Table S12), with four stations having the largest variations (p < 0.01): Abay at Kessie, Chemoga, Temcha, and Uke (Supplementary Figure S1(c)).

Many researchers have examined the effect of LULC changes on the streamflow in the Abay basin. For instance, Mekonnen et al. (2018) explored the combined and individual effects of LULC changes and climate change on the streamflow in the Abay basin; they found that the mean annual streamflow in this basin increased by 16.9% between the 1970 and 2000s.

Worqlul et al. (2015) also stated that by the end of the 21st century, the streamflow in both rivers in the dry seasons would increase by up to 64% but that in the wet season would decrease by 19%. According to Ayele et al. (2023), from 1991 to 2008, the Abay (upper Blue Nile) River basin's flow rate in the wet season increased but that in the dry months decreased. Moreover, Bitew & Kebede (2024) found that LULC changes caused the monthly streamflow in the wet period to increase by 5.81% and that in the dry period to decrease by 3.34%.

This study used the HBV-Light model to examine the effects of LULC on streamflow between 2002 and 2013 while maintaining other parameters constant. The KGE performance for calibration and validation was good having the values of 0.889 and 0.78 for 2002 and 0.82 and 0.766 for 2013. The average simulated streamflow values over the study area under the LULC conditions in 2002 and 2013 were 384.57 and 397.05 mm/year, respectively; however, the average observed streamflow was 315.16 mm/year (Table 4).

The results presented in the aforementioned paragraph indicate when all other parameters were maintained constant, LULC changes caused the average streamflow to increase by 141.69 mm/year according to the HBV-Light model result (Table 5). Table 5 also presents the differences in the simulated streamflow values obtained under the LULC conditions in 2002 and 2013. As presented in this table, the smallest and largest differences between these values (7.15 and 18.63 mm/year) were obtained for 2004 and 2010, respectively.

This study is expected to be extremely significant and plays a vital role in finding the streamflow information of the Abay basin for water resources development, providing evidence on how to make educated judgments and action plans, and serve as an input for researchers and planners in the basin, to help them grasp the changes.

Data scarcity was the limitation of the study. First, even though streamflow data for over half of the basin was required for the analysis of streamflow changes, the data were not available because of data shortages. Second, since there was less dedication to developing the rating curve equation, the recent row data was not converted to discharge. Third, because of concerns about data security and recent gauging station calibration, it was challenging to obtain the long length of measurement period data needed to analyze long-term streamflow variations. Finally, because it is difficult to establish gauging stations in the river due to strong gorges and canyons, certain major tributaries, like the Beshilo River, lack stream measuring stations.

It is recommended that a database be established to access current data in all areas pertinent to carrying out different tasks to conserve the basin as a whole, in addition to making an effort to place gauging stations in suitable locations for the major rivers. The study was restricted to a certain number of stations and a specified measuring period (years) due to the lack of data.

Future studies should examine streamflow variability estimates by investigating additional stations and using extended time-series data. Furthermore, to compare performance with the HBV-Light model, this study does not include additional hydrological models (such as SWAT), which should be included in subsequent studies to increase the accuracy of streamflow variations in the basin.

Understanding the features, long-term patterns, and change points of streamflow for 29 data collected stations in the Abay basin and evaluating the impact of LULC change using the HBV-Light model for the watershed with the largest streamflow changes were the goals of this work for making immediate informed decisions and action plans. To detect the variations effectively, the annual mean streamflow, maximum streamflow, and 7-day average minimum streamflow were analyzed for each station.

The annual 7-day average minimum streamflow of 23 stations (79.31%) considered in this study was less than 1 m³/s, which suggests that nearly all tributaries of the Abay basin are seasonal. MK and Pettit tests were conducted to assess trends and change points, respectively. The trends of mean annual streamflow showed that all of the stations showed either increasing or decreasing values with most stations showing increasing trends. For all stations except Abay at Bahir Dar, an increasing or decreasing trend in annual 7-day average minimum streamflow was discovered.

Moreover, for all stations except Muger, an increasing or decreasing trend in annual maximum streamflow was found. Thus, the Abay basin exhibited streamflow variation in both the wet and dry seasons. The results of this study also revealed that significant variations in annual mean streamflow and annual maximum streamflow were observed at various times for most of the stations. Of all the stations, the Kessie station was selected for modeling because significant trends and change points were discovered for this station.

The HBV-Light hydrological model was used to simulate streamflow values for the Abay basin to know the effect of LULC on streamflow for well-informed decisions. Twelve years (2002–2013) were used to calibrate and validate the HBV-Light model, 1 year for warming up (1 January 2002–31 December 2002), 7 years for calibration (2003–2009), and 4 years for validation (2010–2013). The model had KGE values of 0.889 and 0.780 in the calibration and validation phases, under the LULC 2002 and 2013, respectively, conditions in the Abay basin in 2002. The corresponding values obtained under the LULC conditions in 2013 were 0.820 and 0.766, respectively.

The KGE values obtained with the LULC map for 2013 were smaller than those obtained with the LULC map for 2002 because the model parameters were maximized based on the LULC map for 2002. The average simulated streamflow values over the Abay basin under the LULC conditions in 2002 and 2013 were 384.57 and 397.05 mm/year, respectively; while the average observed streamflow in this basin was 315.16 mm/year. LULC changes were found to result in the average streamflow increasing by 141.69 mm/year when other parameters were maintained constant. The MK result is consistent with the modeling result as streamflow increases.

Overall, the results of this study can be used as a reference by researchers and planners to develop suitable strategies for addressing the growing demands for water from the Abay basin. However, gauging stations must be installed at suitable locations in large rivers so that relevant data can be collected to generate predictions of streamflow variation for the entire basin. Such predictions of streamflow can then be used for the management of the Abay (upper Blue Nile) basin.

The authors thank the Ethiopian Ministry of Water, Irrigation, and Energy and the Abay Basin Authority for providing the data required for this study free of cost. We also thank Wallace Academic Editing for providing the editing service for this paper.

Conceptualization: D.A.K. and C.-C.H. Methodology: C.-C.H. Software: M.A.B. and Y.N.A. Validation: D.A.K. and C.-C.H. Formal analysis: D.A.K. and M.A.B. Investigation: D.A.K. and M.A.B. Resources: C.-C.H., M.A.B., and Y.N.A. Data curation: D.A.K. and Y.N.A. Writing – original draft preparation: D.A.K. Writing – review and editing: C.-C.H. All authors have read and agreed to the published version of the manuscript.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.