Hydrological responses to land use and land cover change and climate dynamics in the Rift Valley Lakes Basin, Ethiopia

Land use and land cover (LULC) and climate change are two main driving forces that lead to large variations in hydrological processes (Savary et al. 2009; Wang & Hejazi 2011; Feng et al. 2014). LULC is the main controlling factor of local hydrological processes and its change can influence both surface water hydrology and soil hydraulic properties (Bormann et al. 2007; Savary et al. 2009; Tigabu et al. 2019; Wagner & Fohrer 2019). Different LULC types have different effects on the runoff processes and the rates of infiltration, erosion, and evapotranspiration (ET) (Baldyga et al. 2008; Hernández-Guzmán et al. 2008), so LULC does not only change the volume of surface and groundwater, but also changes the regional water movement. The expansion of urbanization, agriculture, and deforestation will decrease water retention and infiltration, and increase surface runoff and flood frequency (Wang & Hejazi 2011). In addition, climate change, especially change in precipitation and temperature, greatly affects terrestrial water resources (Cao et al. 2018). Streamflow, one of the most crucial components of the hydrological cycle, can be impacted by climate change both directly via changes in precipitation and indirectly through changes in temperature (Yang et al. 2019). Although climate change is expected to have a negative influence on water resources and freshwater ecosystems in almost every part of the world, the magnitude and nature of this impact vary (IPCC 2007; Wagner & Fohrer 2019; Mahmoodi et al. 2021; Tigabu et al. 2021; Lei et al. 2022). Few studies also indicate that the changing climate pattern will significantly affect East African communities (Bahaga et al. 2015, 2016; Ongoma & Chen 2017; Takele et al. 2020). For instance, there has been an increase of 1.5 °C of the daily maximum temperature in the central Rift Valley basin over a period of 37 years (Jansen et al. 2007), and associated ET is also expected to increase in the range of 3–4% (Gadissa et al. 2019). This will result in water loss in the basin and the effect will further be exacerbated if the temperature continues to increase. The effects of land cover and climate changes on the runoff of a basin are of great interest to water resource planners, managers, and decision-makers (Helmschrot & Flügel 2002). However, many water resources impact studies, particularly in the Rift Valley Lakes basin, have focused on the long-term evaluation of the basin's water balance due to climate change only (Alemayehu et al. 2006; Legesse & Ayenew 2006; Awulachew et al. 2007; Jansen et al. 2007; Ayenew & Tilahun 2008; Raju & Kumar 2018). Even though LULC change is taking place at a rapid rate, and deforestation activities are having a major impact on water resources, analysis of its effect on hydrological processes remains limited in the basin. Several recent studies have estimated the impact of LULC on hydrological processes in various parts of the world (Hernandez et al. 2000a; Ma et al. 2009; Dile et al. 2013; Yang et al. 2014); few studies have investigated the combined effect of climate and LULC change on hydrological processes (Hernandez et al. 2000a; Jingjing et al. 2017; Zhao et al. 2018), and on streamflow (Roderick et al. 2014; Pan et al. 2017). As changes in water resources also feedback on changes in LULC (Wagner & Fohrer 2019), quantitative and combined assessment of climate and LULC change impact on hydrological processes is important for the sustainable development of land and water resources.

There are three fundamental types of approaches for quantifying the impact of LULC and climate change on streamflow: hydrological modeling, sensitivity analysis, and time series analysis (empirical statistical approach) (Li et al. 2022). The empirical statistical approach is mainly based on trend and correlation analysis of time series data (Ahn & Merwade 2014), which are easy to apply but they require a long-term time series of hydro-climatic data (Jiang & Wang 2016). This type of method may not be able to capture the exact physical characteristic of the hydrological system (Liu et al. 2012). Sensitivity analysis is mainly based on the water balance components within a basin and can be applied to streamflow (Zhang et al. 2015). This method allows estimation of the sensitivity of the streamflow to climate change using sensitivity indices with few data (Wang et al. 2012; Tu et al. 2015), but cannot directly quantify the effects of LULC, which limits its value (Asenso Barnieh et al. 2020). Hydrological modeling approaches usually divide the whole study period into a base period and an interference period, and the simulation results for these periods are then compared to quantitatively distinguish the effects of climate change and human activities on streamflow. Models need to consider a variety of parameters, and considerable efforts to establish a suitable model for analysis (Tan et al. 2020).

Previous studies have associated streamflow change with high abstraction of water for irrigation and industries (Zinabu & Elias 1989; Kebede et al. 1994; Ayenew 2002), climate change (Alemayehu et al. 2006; Jansen et al. 2007; Van Halsema et al. 2011; Belete et al. 2015), environmental degradation (Ayenew 2004), volcano-tectonics and sedimentation (Street & Grove 1979; Le Turdu et al. 1999), and frequent earthquakes (faults) (Ayalew et al. 2004; Belay 2009). These studies focused on single-driven impact analysis and do not quantify the relative impact contribution of these drivers on streamflow. However, the rainfall-runoff relations within a watershed are mainly driven by the interplay of climate, LULC, and soil (Hernandez et al. 2000b), and understanding these impacts requires an accurate assessment of the different conditions for a given hydrological system. Therefore, an integrated hydrologic model is used to quantify the relative contribution of climate and LULC change for the basin region under various scenarios so that a deeper understanding of the relative and combined influence of climate and LULC change can be obtained. The main objective of this research is to understand and evaluate the hydrological response to climate change and LULC dynamics. We hypothesize that the combined impact of climate and LULC dynamics will be stronger in changing water fluxes than the isolated impacts.

The Great Rift Valley is a 4,000 km long fault line that stretches from the Red Sea to Mozambique's and Zambia's Valley. It is the single largest geographical feature in Africa and it transects Ethiopia, Kenya, Uganda, Rwanda, Burundi, Zambia, Tanzania, Malawi, and Mozambique (Gregory 2018). The Rift Valley Lakes basin is a central part of the Great Rift Valley, which is one of the most important basins in Ethiopia, occupying an area of 55,050 km².

The climate of the basin is heterogeneous in space and time. With arid, semi-arid, humid, and semi-humid sub-basins, it is characterized by low rainfall, high temperature, and high ET. The basin is characterized by three climate regions. The Abijata-Ziway sub-basin which is characterized by low annual average rainfall ranging from 400 to 860 mm with one peak in August. The second basin, the Abaya-Chamo sub-basin, is characterized by relatively higher annual precipitation ranging from 704 to 1,200 mm with a bimodal climate pattern with peaks in April and August. The Segen sub-basin is characterized by a bimodal rainfall pattern with peaks in April and October and receives an average annual precipitation ranging from 500 to 1,100 mm. The temperature varies throughout the basin and ranges from 10 to 36 °C, with the warmest area on the Rift floor to frost-prone area in the Afro-alpine zone. The southern part of the Rift is lower in elevation, warmer, and drier than the other parts of the basin.

The Ethiopian Rift Valley Lakes basin consists of a chain of lakes, streams, springs, and wetlands with unique hydrological characteristics. Based on the digital elevation model (United States Geological Survey) data, the slope ranges from 0 to 161%. The slope class smaller than 2% accounts for about 7.1%, the slope class within the range of 2–4% accounts for about 19.5%, the slope class within the range of 4–8% accounts for about 72.0%, and the remaining 1.4% has a slope range from 8 to 161% from the total area. Agricultural land is the most dominant land use followed by vegetation and water body.

The main hydrological characteristics of the basin have been acquired from the climate, topography, land use, and soil properties. Time series of climate data for 44 stations (1981–2018), LULC map of 1989 and 2009 in previous research (Ayalew et al. 2022), streamflow (1983–2000), the Shuttle Radar Topography Mission digital elevation model (SRTM DEM) (30 × 30), and soil data were used for this study. The SRTM DEM has been used, from which slope, river network, and watershed boundaries are obtained. Selected soil physical properties and the area coverage of each soil type were classified based on the requirements of the SWAT+ model (Table 1).

The hydrological system and processes of this basin are modeled using the Soil and Water Assessment Tool Plus (SWAT+), a completely revised version of the SWAT model. It is more flexible than SWAT in terms of the spatial representation of interactions and processes within a watershed (Bieger et al. 2017). Some algorithms are used to calculate the processes, the structure and organization of the code, and the input and output files, and some parameters have been modified in the model (Bieger et al. 2017).

SWAT+ simulates the hydrological processes from precipitation to streamflow using the water balance equation. The algorithm simulates a land and a water phase, in which the model simulates soil water content, surface runoff, ET, infiltration, percolation, and return flow on a daily basis (Moriasi et al. 2015). The variable storage routing method was used for river routing and the Soil Conservation Service curve number infiltration method was used. The Hargreaves estimation equation was also used to calculate the potential ET. The Hargreaves method is a potential ET estimator in data scarce regions (de Sousa Lima et al. 2013; Moeletsi et al. 2013). It can estimate ET using only minimum temperatures and maximum temperatures.

Apart from the spatial scale, depending on the concern of the analysis, SWAT+ simulates daily time scales. Each term of the water balance equation has detailed physical processes that are interlinked in harmony related to the atmosphere–vegetation–soil consortium. Further details of the model are available at http://swatmodel.tamu.edu and literature (Bieger et al. 2017; Arnold et al. 2018; Bieger et al. 2019).

Currently, SWAT+ is tested in few watersheds across the world, where the results from SWAT+ are favorable compared to the previous model version (Wagner et al. 2022), and SWAT+ is also capable of simulating streamflow (Arnold et al. 2018; Bieger et al. 2019; Wu et al. 2020). It can simulate the quantity and quality of surface and groundwater resources from a hydrological response unit to a basin scale. It is also suitable and capable of assessing the long-term impact of climate change, land use change, and watershed management practices on water resources. It characterizes the spatial heterogeneity of a watershed. Each watershed is divided into multiple homogenous hydrological response units based on a unique combination of land use, slope, and soil characteristics Hydrological Response Unit (HRU) (Gassman et al. 2007). This spatial unit is assumed to respond similarly to hydrological inputs in SWAT+. In this study, the hydrological processes are evaluated from the outputs of HRU and watershed levels.

In the current research, local sensitivity analysis/Morris' screening (one factor at a time (OAT)) was adopted for 18 parameters from which 10 were found to be sensitive for streamflow in five watersheds (Table 2). Two of the watersheds (Katar and Meki) are from the monomodal rainfall region (R I) and three of the watersheds (Binate, Gidabo, and Upper Gelana) belong to the bimodal rainfall region (R II). We found the same sensitive parameters for all watersheds, but different ranges due to different catchment characteristics and rainfall patterns. The largest changes in response to parameter perturbations were selected by combining the Kling–Gupta efficiency (KGE) (Gupta et al. 2009), the Nash–Sutcliffe efficiency (NSE) (Nash & Sutcliffe 1970), percent bias (PBIAS), and root mean square error (RMSE)-observations standard deviation ratio (RSR).

The streamflow data at the watershed outlet (at the gauges) from 1980 to 2000 was split into three periods for model warm-up (1980–1982), model calibration (1983–1994), and validation (1995–2000). For calibration, 5,000 parameter sets were generated by Latin Hypercube Sampling (Soetaert & Petzoldt 2010) using the parameter ranges given in Table 2. The model configurations were evaluated for the same 5,000 parameter sets. For each parameter set, a model run was performed and the final parameter set was selected based on the best combined KGE, the NSE, and PBIAS values so that the KGE, NSE, and PBIAS were used as an objective function in this research. The model was also examined during the validation period with the same parameter set. Calibration and validation were carried out in R using the packages FME for Latin Hypercube Sampling (Soetaert & Petzoldt 2010), hydroGOF for model evaluation (Zambrano-Bigiarini 2014), and the packages zoo (Zeileis & Grothendieck 2005) and xts (Ryan & Ulrich 2011) for data processing.

In addition to statistical performance evaluation, visual inspection of the hydrography and flow duration curve (FDC) could provide information about the overall qualitative match between measured and simulated streamflow. The performance evaluation is based on the daily temporal scale.

The annual average water balance analysis does not reflect the inter-annual variability of water balance components. The flow signals are averaged on inter-annual scale and it is difficult to distinguish which flow signals are impacted in the annual water balance analysis. To avoid this limitation, we used FDC to evaluate the seasonal variation of streamflow for the baseline and interference periods. The shape of the curve is an index of the natural storage in the watershed, including the groundwater. The dry season flow consists entirely of return flow from the groundwater, i.e., the lower end of the FDC indicates the general characteristics of shallow aquifers; therefore, flow signals were determined from the FDC based on the exceedance probability threshold. Here, low flow signals are defined as ≥ 75% of the exceedance probability, while higher flow signals are defined as ≤ 20% of the exceedance probability and the rest are mid flow. The choices of thresholds for disaggregation were based on earlier studies (Yilmaz et al. 2008; Pfannerstill et al. 2014).

The land use has been classified using pixel-based ensemble machine learning algorithms Random Forest on R environment for the years 1989 and 2009 in previous research (Ayalew et al. 2022). The spatio-temporal land use dynamics have been quantified and compared with the help of dynamics and transition matrix indexes (Singh et al. 2017).

To evaluate the impacts of climate change and LULC on hydrological processes, scenario combination is often used when simulating the water balance components within the hydrological models (Guo et al. 2014). The result are shown in Subsection 3.5, the baseline period (1980–2000) and the interference period (2001–2018) can be determined, and then the scenarios for quantitative attribution analysis can be set as shown in Table 3.

S1 is the scenario regarded as the reference scenario, which uses the LULC and climate data in the baseline period. By comparing the simulation results under S1 and other scenarios (S2, S3, and S4), the contributions of climate changes and LULC to water balance components (e.g., the runoff, lateral flow, percolation, and ET) in a watershed have been quantified. The LULC data in the baseline period and climate data in the interference period are used to quantify the contribution of climate change (S1 and S2). Furthermore, to quantify the contribution of LULC, S3 is designed using the LULC data in the interference period and climate data in the baseline period. Also, S4 is designed to quantify the combined impacts of both climate change and LULC using the LULC and climate data in the interference period.

Model calibration and validation have been carried out for daily streamflow following the sensitivity analysis. The performance of the model runs for each watershed was evaluated using objective functions NSE, PBIAS, KGE, PBIAS, and RSR (Table 4). The statistical indices of the model showed that the model has strong predictive capability in all watersheds during both calibration and validation periods.

Table 4

Daily and annual average model performances

ΔS (Precip is precipitation, Qusur is surface flow, Qlat is lateral flow, Perco is percolation and ET is evapotransportation) is a change in storage.

View Large

The SWAT+ model resulted in the annual average ET ranges from 629.1 to 1.053.0 mm; and other research showed also that the ET is very high over the basin (e.g., 985.5 mm/a (Meaza et al. 2019), 958.3 mm/a (Daniel & Abate 2022), and 900 mm/a (Billi & Caparrini 2006). As a confirmation, the observed annual evaporation data from 2000 to 2018 was obtained from Arba Minch meteorological station and it shows that the evaporation ranges from 630 to 747 mm. This indicates that the model result is reliable.

However, the flow duration curve segment from 20 to 40% exceedance probability indicated an overestimation of streamflow by the model, especially in the Meki, Bilate, and Gelana watersheds during the validation period. The model also underestimates high flows and low flows during the validation period in all watersheds. Furthermore, as indicated in Table 2, the best model run was obtained with the base flow recession constant, ALPHA_BF ≤ 0.3. This indicates that the groundwater flow response to changes in recharge is slow in the basin. The value of PERCO is also ≤ 0.5 which indicates that the water leaves the bottom of the root zone quickly and reaches the shallow aquifer. The majority of the soil in the basin is sand and sandy loam which favor a fast percolation and groundwater recharge. The hydrologic response in the lateral flow is also fast, as values of LATQ_CO are <= 0.3.

In addition, the overall temporal dynamics of land conversion (‘from–to’) have been evaluated as presented in Table 5. The LULC conversion matrix indicates that most land uses have changed to agricultural land. It is worth mentioning that the maximum land transformation was observed in the wetland as only 2.3% of its original cover remained unchanged. Water, alpine, agriculture, and settlement have relatively high shares of persistent land use (>65%). Since different LULC types have different impacts on water balance components (like ET, surface runoff, lateral flow, and percolation) of a watershed, these high transformations of LULC over the basin would have a high effect on the water balance components.

To evaluate the individual impact of LULC and climate dynamics on hydrological processes, we used the LULC and climate dynamics results from the previous research (Ayalew et al. 2022). The dynamic nature of rainfall and temperature is different from 1998 to 2000 and 2001 to 2018; and combining the change point tests for precipitation and temperature, the middle year 2000 was regarded as the change point from 1980 to 2018. Therefore, in this study, years before 2000 were classified as the baseline period and years after 2000 were classified as the interference period. Then, the temporal trends of precipitation and temperature during these two periods were obtained using Mann–Kendall trend test. Distinct patterns were observed for these two climatic factors starting from 1980, with temperature exhibiting significant annual change, while precipitation showed insignificant variations. (Table 6).

Precipitation showed an insignificant decreasing trend in both periods, with a P-value of the baseline period (ranging from 0.381 to 0.695) and the interference period (ranging from 0.383 to 0.60), while temperature records had a significantly strong increasing tendency for the interference period and a significant increasing trend for the baseline period with a P-value of 0.00036 and 0.0144 with regard to the 5% significance level standard, respectively.

Furthermore, the change of spatio-temporal distribution of surface runoff, lateral flow, and percolation varied across the basin, with the annual average change value ranging from 7.8 to 13.14%, −4.47 to −16.51%, and −3.28 to −10.19%, respectively. High variation was observed in Bilate, Gelana, and Gidabo watersheds. The variation of change of spatio-temporal distribution indicates that different watersheds have diverse runoff generating capacities.

Table 7 shows the evaluation of the LULC, climate dynamics, and combined impact on the water balance components during the study period. The results showed that change in LULC had a positive impact on surface runoff while having a negative impact on infiltration and ET. Although climate dynamics positively influenced ET in all watersheds, and had a negative impact on surface runoff and infiltration. LULC had a greater relative contribution to changing surface runoff than the combined and climate impact, whereas the relative contribution of combined impact was greater in the changing of ET and infiltration.

During the study period, the vegetation area decreased by 20.7%, the water bodies decreased by 3%, the agricultural land increased by 23.14%; the barren land increased by 0.13%; the grassland increased by 0.11%; and the settlement area increased by 0.32% from the total share of land. This high reduction of natural vegetation cover and rapid increase of agricultural land is mainly associated with rapid population growth and agricultural expansion land use policy. In addition, harvesting of trees for firewood and charcoal is another important activity reducing tree density of vegetation cover. Especially, the National Economic Development Strategy (NEDS) emphasizes the need for the agricultural sector to enhance food self-sufficiency, ensure food security at the household level, and for agriculturally-led industrial development-aggravating deforestation and agricultural expansion. Apart from population growth and national land policy, high inter-annual climate variability has impacts on land use dynamics especially on water and wetland. The surface area of water and wetland has decreased noticeably in the land use map of the interference period, which is mainly associated with a reduction of annual rainfall due to frequent drought occurrence and water abstraction, e.g., for irrigation and industrial water use (Ayenew & Tilahun 2008).

Apart from high inter-annual climate variability, the annual average rainfall decreased in the interference period. This reduction of rainfall was mainly associated with a higher frequency of drought occurrence in the interference period than in the baseline period. The spatio-temporal variations of rainfall in the Rift Valley Lakes basin mainly depend on the north–southward movement of the Intertropical Convergence Zone (Korecha & Barnston 2007; Diro et al. 2009) and its magnitude is affected by the variation of equatorial Pacific sea surface temperature (Gissila et al. 2004; Block & Rajagopalan 2007; Diro et al. 2011). The detailed climate characteristics of the basin have been explained (Ayalew et al. 2022).

Meanwhile, the decline of annual average ET in the watersheds is a result of LULC change, and climate change is mainly attributed to the expansion of agricultural land at the cost of natural vegetation. Especially, the magnitude of change in ET is higher in the southern part of the basin, which was covered with vegetation in the baseline period. Elsewhere, studies have shown that actual ET is generally greater for vegetated land than non-vegetated land, which is attributed to the loss of transpiration and to the reduction in soil moisture in non-vegetated LULC types due to the loss of vegetation cover (Chen et al. 2013; Tigabu et al. 2021). Moreover, forests have a much higher transpiration than other vegetation types, leading to a higher annual ET value (Yang et al. 2012).

The decrease in rainfall and the increase in temperature in the basin not only lead to the change of ET, but also a change of infiltration and streamflow. Meanwhile, climate and LULC dynamics not only affect ET, but also, they could affect surface runoff and infiltration. The increase in surface runoff is primarily due to deforestation and urbanization, which results in reduced infiltration. During the interference period, high deforestation resulted in greater surface runoff and lower infiltration. Urbanization leads to an increase in impervious areas, resulting in decreased soil infiltration. With the reduction in base flow contribution to streamflow, the surface runoff increases, which can lead to more recurrent and severe flooding (Rose & Peters 2001; Massei & Fournier 2012; Ehret et al. 2014). The increase in streamflow is also attributed to a decrease in water bodies and forest cover. It is worth mentioning that our study investigated a reduction in annual average ET caused by LULC change that was relatively larger in magnitude than the offsetting increase caused by climate variability, indicating that the LULC change played a dominant role in affecting annual average ET.

The internal variability of streamflow was also evaluated using FDC. The results indicate an increase in the frequency of both high and low flows. The increase of high flow during the wet season was connected with deforestation and urbanization, whereas the increasing low flow frequency during dry season can be linked to irrigation abstraction (Wendemeneh et al. 2020), frequent drought, and a decrease in infiltration.

LULC and climate are significant environmental factors that influence hydrological processes throughout a watershed. The present study attempted to understand and investigate the contribution of existing dynamics of climate and LULC on the hydrological processes of the Rift Valley Lakes basin. Main water balance components were simulated using a semi-distributed SWAT+ model. The SWAT+ model performed well for daily streamflow during calibration and validation periods.

The change of the main water balance components in the Rift Valley Lakes basin was related to both climate and LULC dynamics during the study period. Four scenarios were designed using combinations of two periods of climatic data (1980–2000 and 2000–2018) and two sets of land use maps (1989 and 2009).

This research investigates the individual and combined impact of land use change and climatic variations in the watersheds using a semi-distributed SWAT+ model. The results affirmed that the individual impact of LULC change contributes to an increase in the surface runoff and a decrease in ET and infiltration primarily due to an increase in deforestation, urbanization, and a decrease in water bodies of the basin. The combined impact of climatic variations and land use change was higher on ET and infiltration, but the isolated impact of LULC was higher on surface runoff. Generally, our result depicted that LULC change and climate dynamics can modify surface runoff, infiltration, and ET in the Rift Valley basin, and that surface runoff, infiltration, and ET are important hydrological components subject to change with LULC and climate dynamics. As the changes in LULC and climate variability in the basin continue to increase, further significant change can be expected in the hydrological components. Thus, further investigations of the hydrological responses under future LULC and climate change scenarios are important for future sustainable land and water management strategies. The study provides insight into the hydrological response to dynamics in climate and land use changes in the Rift Valley basin in recent decades. The results of this study can be beneficial to the authorities, decision-makers, water resource engineers, and planners for the best water resource management approaches in the perspective of LULC change and climate change of a heterogeneous climate region such as that of the Rift Valley basin. This method can be applied to other regions, and a comprehensive understanding of future scenarios of LULC and climate change should be combined. By doing so, a more holistic approach can be achieved.

The German Academic Exchange Service (DAAD) Program of Germany and the Ethiopian Ministry of Education supported this research. Our grateful acknowledgment is also toward the Ethiopian Ministry of Water Resources and Energy, Ethiopian Meteorology Agency for providing hydro-climatic datasets.

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

The authors declare there is no conflict.