Quantification of the hydrological consequences of climate change in a typical West African catchment using flow duration curves

Precipitation and temperature variables are an integral part of the hydrological cycle; therefore, any changes that occur in the climate might also affect the hydrological regime, and there is a direct connection between the climate and the hydrology of an area. Several studies and reports (Jung & Kunstmann 2007; Solomon 2007; Wagner 2008; Speth et al. 2010; Patricola & Cook 2011; Bárdossy & Pegram 2012; Nikulin et al. 2012; Stocker et al. 2013; Laprise et al. 2013; Ibrahim et al. 2014) have projected that climate change (CC) has been happening and will continue to happen in the future. Changes in a river's ecohydrological processes may have consequences for riparian countries whose livelihoods are derived from the hydro system (Kohler et al. 2015).

West Africa is a fast developing subregion of the African continent which depends largely on surface waters for its hydropower generation. Rain-fed agriculture is also the main source of occupation and livelihood in the region. The major water resource usage in the region includes hydropower, small-scale irrigation, and domestic water supply. Any slight changes in the hydrology of the basins in the subregion will significantly affect the citizens and the economy. The domestic water supply in West Africa has also been threatened because of changes in rainfall patterns and the drying-up of some of the major rivers. Although population increases, pollution, and land-use changes could be factors for these problems, one cannot rule out the adverse effects of CC on hydrology and water resources.

Huge investments have been made in hydropower generation in this subregion over the past one to two decades but have yielded small results in terms of electricity production. An example is the Bui hydropower in Ghana which is working far below the designed capacity because of low levels of water in the reservoir of the dam. For the rapid development of the African Continent and its subregions, agriculture, water supply, and energy production are key to the realization of this dream. Quantifying the future hydrological regime in the subregion, taking into consideration CC, would be very instrumental in helping policymakers and water resource managers to make important decisions on investment, development, and management.

Studies by Kunstmann & Jung (2005), Jung (2006), Jung & Kunstmann (2007), and Jung et al. (2012) on the effects of CC on the hydrology of the Volta basin in Western Africa showed a very heterogeneous response of river runoff to changes in climate variables. Their work showed that there was no major change for the mean runoff regime of the future as compared with the current simulation.

Roudier et al. (2014) summarized the state of knowledge on potential future streamflow evolutions given global warming in West Africa by creating a quantitative database of 301 points of information on the impact of CC on runoff in the region. The results showed no clear general pattern of future runoff evolution in this area. The median of the distribution was 0% with a mean of +5.2%, after considering different time horizons, different rivers, scenarios, and models. The correlation analysis they performed revealed that runoff changes are tightly linked to changes in rainfall (R = 0.49). Kankam-Yeboah et al. (2013), however, projected a significant decrease in the projected mean annual streamflow for the years 2020 and 2050 for the White Volta and Pra Basins. Similar results of a downward trend of streamflow in the future for the West Africa subregion were obtained by Giertz et al. (2010) and McCartney et al. (2012). Götzinger (2007) reported similar results of a considerably reduced discharge, especially in the rainy season due to reduced precipitation and increased temperature for two different climate scenarios which were simulated for a basin in Benin (West Africa) in comparison to a simulated historical period. To improve the acceptance of these results by society and stakeholders, the continual update of the results with new models and scenarios is recommended (Götzinger 2007).

According to Vogel & Fennessey (1995), flow duration curves (FDCs) typify the old Chinese proverb ‘one picture is worth a thousand words’ through their ability to summarize a wealth of hydrologic information into a single graphic image and it could be used in summarizing the results of detailed and complex water resource studies. They are often used to summarize the impacts of potential CC scenarios on water resource systems (Schwarz 1977). The FDC shows the relationship between the magnitude and the frequency of daily streamflow for a river basin, and it provides an estimate of the percentage of time the streamflow was equaled or exceeded over a historical period. The majority of studies in FDCs are about the utilization domains of duration curves in water resources engineering (Cigizoglu & Bayazit 2000). They have been recommended for use in hydrological studies such as water supply, hydropower, waste-water treatment plant capacity, irrigation, and planning (Male & Ogawa 1984; Warnick 1984; Cigizoglu 1997). An approach based on the FDC for patching and extending observed time series of daily streamflow was developed by Hughes & Smakhtin (1996). Regional FDC procedures have been developed in different regions of the USA for ungauged sites by Fennessey & Vogel (1990) and Smakhtin et al. (1997). Although FDCs provide a convenient, simple, and powerful tool for studying the flow regime characteristics in a river basin under climate and land-use change (Pumo et al. 2013), the number of the studies related to CC impacts on hydrology (Arora & Boer 2001; Sellami et al. 2016; Langat et al. 2019) is limited.

Previous studies (Kunstmann & Jung 2005; Jung 2006; Jung & Kunstmann 2007; Solomon 2007; Wagner 2008; Biasutti et al. 2009) on the topic in the subregion had focused mainly on monthly to seasonal changes of temporal resolutions with shorter durations (i.e. 10 years), maybe because of the computational run-time demands of the models they used. In this study, we use a daily temporal resolution for the simulation and analysis and extended the simulation and climate periods to 45 years.

The objective of the study is to use simple, less data demanding, but robust tools such as the Hydrologiska Byrans Vattenbalansavdelning (HBV) hydrological model and FDCs are used to try and quantify the hydrological consequences of CC in the subregion after performing bias correction on the climate simulations. The bias correction of the various climate simulations used for the study was done using two different methods. The downscaled and corrected climate simulations were used in conjunction with a calibrated and validated HBV hydrological model to determine the consequences of CC on the hydrology of the Black Volta Basin. The study contributes to the knowledge of CC impacts and also provides a simple, rapid, and powerful tool for assisting water stakeholders in identifying the flow regime components and formulating mitigation and management strategies. The information gathered may be useful in the development of new sustainable water-related projects and river basin management systems.

In this study, general circulation model (GCM)/regional climate model (RCM) climate data from the Coordinated Regional Downscaling Experiment (CORDEX) and the Regional Climate Division of the Institute of Meteorology and Climate research at Karlsruhe Institute of Technology (IMK-IFU) were used to determine CC in the West Africa subregion. The study area is the Black Volta Basin in Western Africa with a catchment size of about 155,000 km² (Figure 1) and is shared by four other West African countries (Ghana, Mali, Burkina Faso, and Ivory Coast). The Black Volta River is the main river in the catchment which empties into Volta Lake.

The GCM driving models used for the study are (1) CanESM2M: the second edition of the climate model for the Canadian Center for Climate Modelling and Analysis (Chylek et al. 2011) and (2) EC-EARTH: an earth system model (ESM) developed and maintained by a consortium of European climate modelers, bringing together 27 research institutes from 10 European countries to collaborate on the development of an ESM (Hazeleger et al. 2012), (3) ECHAM: an atmospheric GCM developed by the Max Planck Institute (MPI) for Meteorology, forming the atmospheric component of the MPI–ESM (Stevens et al. 2013), (4) GFDL: climate and ESM developed by the Geophysical Fluid Dynamics Laboratory (GFDL) which is a laboratory in the National Oceanic and Atmospheric Administration (NOAA) Office of Oceanic and Atmospheric Research (OAR) (Dunne et al. 2012), and (5) MIROC: a climate model developed by the Center for Climate System Research in Japan (Watanabe et al. 2010).

The Rossby Centre Regional Atmospheric Climate Model (RCA4) (Kupiainen et al. 2011) and the Weather Research and Forecasting (WRF) (Skamarock et al. 2008) Model were used as the RCM for the data from CORDEX and IMK-IFU, respectively. The analysis was performed on a daily temporal resolution and a spatial resolution of 50 and 12 km, respectively, for the data from CORDEX and IMK-IFU. The GCM/RCM climate simulations often show the bias representation of observed climate (Varis et al. 2004; Christensen et al. 2008; Teutschbein & Seibert 2010, 2012).

There are several bias adjustment methods which range from simple scaling methods to sophisticated approaches employing probability mapping or weather generators. The procedures employ a transformation algorithm for adjusting the RCM output.

In this study, the (multiple) linear-scaling (MC) method and the distribution mapping approach, which is also referred to as ‘probability mapping’ (Ines & Hansen 2006; Block et al. 2009), ‘quantile–quantile mapping’ (Boé et al. 2007; Bárdossy & Pegram 2011; Johnson & Sharma 2011; Sun et al. 2011; Pegram & Bárdossy 2013), ‘statistical downscaling’ (Piani et al. 2010), and ‘histogram equalization’ (Sennikovs & Bethers 2009; Rojas et al. 2011), were used to bias adjust the RCM outputs. The measured rainfall and temperature at the stations (Figure 1) were used as the observed or ‘correct’ climate variables and the bias correction is done for the closest grid points to stations. A precipitation threshold (for a day to be considered wet, precipitation should be greater than 0.1 mm) as used by Piani et al. (2010) was also used to adjust the wet-day frequencies of precipitation time series. ‘MC’ and ‘QC’ are used to represent the linear-scaling and quantile-quantile mapping methods of bias-correction respectively. The two methods are described in the subsequent subsections.

It should be noted that the applied correction/adjustment factors and terms are assumed to remain constant even for future conditions.

Quantile–quantile mapping is to correct the distribution function of RCM-simulated climate values to agree with the observed distribution function and it is done by the creation of a transfer function to shift the occurrence distributions of precipitation and temperature (Sennikovs & Bethers 2009).

The shape and scale parameters for each of the distributions were determined using the maximum likelihood estimation.

The consequences of the changing climate on the hydrology of the region were performed using a validated HBV hydrological model and the UC and corrected climate simulations. The hydrological model performed well for the study area with an average Nash–Sutcliffe (NS) of 0.75 and 0.57 for calibration and validation, respectively. In all, results showed that the model was stable and reacted well to precipitation signals (Kwakye & Bárdossy 2020). Additional information on the application of the HBV hydrological model in the study area can be found in our earlier study (Kwakye & Bárdossy 2020). The observed hydro-meteorological data used for the study were obtained from the Volta Basin Geoportal, Meteorological agencies of both Ghana and Burkina Faso, and from the Hydrological Services Department of Ghana. The duration of the time series is from 1961 to 2005 (45 years) with a daily temporal resolution. The areal precipitation, temperature, and potential evapotranspiration were used as inputs for the lumped HBV hydrological model to get the river flow (discharge) as the output for the historical (control) and future (rcp) time periods. FDCs and annual cycles were developed for the various driving models to quantify the hydrological consequences. A flowchart of the methodology is provided in Figure 2.

The FDC is formulated as F = (R/(n + 1)) (100), where F is the frequency of occurrence expressed as percentage of the time a particular streamflow value is equaled or exceeded (exceedance probability), R is the rank corresponding to streamflow, Q and n is equal to the total number of days. To characterize the information in the FDC, it could be partitioned into different segments: (1) the first part representing high flows (0–10% exceedance probabilities of flow) characterizing watershed water yield from high rainfall events; (2) wet conditions (10–40%) described by flows from medium size rainfall events; and (3) mid-range flows (40–60%) represented by flows from moderate rainfall events. The other segments were dry period conditions (60–90%) related to the intermediate base flow relaxation response of the watershed water yield, and the low flows (90–100%) relating to the sustainability of streamflow in the long term (Langat et al. 2019).

Supplementary Appendix B indicates the performance of the two different bias correction methods applied on the different GCM driving models used for this study. These figures show the annual cycle of areal rainfall for both methods with their UC data for the various driving models.

For precipitation change signal, the CanESM model does not show any clear change. The EC-EARTH, GFDL, and MIROC driving models predict a slight positive change signal in both the UC and bias-corrected GCM/RCM simulations. On the other hand, the ECHAM model indicates a slight decrease in the future precipitation in both the UC and bias-corrected simulations. The two correction methods gave similar precipitation change signals for all the GCM/RCM models used in this study. It could therefore be stated that the future precipitation change signal is difficult to predict for the subregion.

Both the linear-scaling MC and quantile mapping (QC) approaches tried to bring the historical simulations to the observed while maintaining the change signals between the present and the future in most of the models, but the MC method destroys part of the distribution in the UC data for some of the models in the process. The curve from the corrected historical data from the MC method was closer to the observed than the QC in the case of EC-EARTH and MIROC, but for the other driving models, the curve from the adjusted historical data from the QC was closer to the observed (Supplementary material, Appendix B).

The FDC in Figure 3 shows that for the UC RCM simulations, the discharge (river flow) in the future will slightly increase, with magnitudes slightly higher for mean daily flow which is less than 100 m³/s. For the flows which are higher than 100 m³/s, the increase is very minimal. However, the curve of the historical discharge simulations was quite different from the observed discharge curve and the reasons might be from the errors in observations and wrong parameterization of the climate and/or the hydrological models. It could also be seen from the curve that for 25% of the time, the daily discharge obtained from the RCM simulations equaled or exceeded 100 m³/s. This number reduces to about 10–20 m³/s for the 50% exceedance probability and reduces further to 2–3 m³/s for the 75% exceedance probability. This shows that the rivers in this catchment virtually dry up about 25–30% of the time. The peak discharge is about 300 m³/s and this is equaled or exceeded for about 10% of the time.

Figure 3 also shows the results from the bias-corrected simulations using the QC method. They are shown in dashed blue and dashed green colours for historical (1980–2005) and future (2020–2045) periods, respectively. The corrected data predict an increase in the future discharge (river flow) as compared with the historical with the positive signal more pronounced in the corrected data. For example, for 50% of the time, a discharge of 3 and 12 m³/s is equaled or exceeded in the corrected historical and future, respectively. The change signal is reduced in higher discharges (i.e. flows greater than 100 m³/s). In all three cases (i.e. the UC, corrected using QC, and corrected using MC), the change signal is maintained and reproduced. The simulated discharge curve (sim) is the simulation obtained by using the observed climate as input in the hydrological model to produce the runoff (discharge). This curve is close to the observed discharge but it underestimates the low flows and overestimates the high flows (Figure 3). A possible explanation of the under-/overestimation of the flows could be because models are imperfect representations of the real world and thus model uncertainties will always be present. Inherent complexities of natural mechanisms, as well as inappropriate assumptions within the entire modeling procedure, give rise to these uncertainties.

The discharge annual cycle from the RCM simulations using the CanESM driving model showed a bimodal discharge cycle for the UC and corrected data using QC with the first peak in May and the other in August. The observed annual cycle and the corrected data using the MC method showed a unimodal discharge cycle which peaks in August/September (Figure 4). From the RCM simulations, there is a decrease in the discharge for the first cycle and an increase during the second cycle for the future as compared with the historical simulation. It could be inferred that there will be a decreased discharge in the first/minor discharge cycle and an increased discharge during the second/major discharge cycle in the future. The rivers dry up for about 4 months of the year. Both the UC and corrected data show the discharge cycle reaching its highest amount in August as in the case of the observed discharge.

The discharge from the historical RCM simulations did not quite represent the observed discharge with the simulated discharge (sim) outperforming the discharge resulting from the historical RCM simulations. While the simulated discharge showed a unimodal annual cycle which is what happens in the observed data, the historical showed a bimodal cycle and there is a slight underestimation of the low flows and overestimation of the peaks in the simulated discharge (Figure 4).

The FDC from this driving model reveals that the discharge will decrease in the future as compared with the present which is in contrast to what the CanESM model predicted. In the first segment representing high flows (0–10% exceedance probabilities), the FDC shows a slight increase in the future flow characterizing a higher watershed water yield from high rainfall events in the future. The other segments of the curve (10–100% exceedance probabilities) for this driving model representing wet conditions (10–40%), mid-range flows (40–60%), dry period conditions (60–90%), and low flows (90–100%) all show a decrease in the future flows. Most worrying is the reduction of the low flows (90–100% exceedance probabilities) in the future relating to the sustainability of streamflow in the long term, which could lead to the drying-up of the watershed (Figure 5). For 50% of the time, a discharge of 18 and 28 m³/s is equaled or exceeded for the future and historical periods, respectively (Figure 5).

The bias-corrected data also confirmed a negative signal for future discharge. At 25% of the time, a discharge of 70 m³/s is equaled or exceeded, while 20 m³/s is for 50% exceedance probability and 4 m³/s for 75% in the future. The future flow for 50% exceedance probability is 20 and 8 m³/s for QC and MC methods of bias correction, respectively (Figure 5).

The annual cycle of discharge (Figure 6) also confirms a slight decrease in the future flows and shows one discharge cycle and not two as seen in the case of the CanESM model. The flow begins to increase in April when the rainy season starts and peaks in August/September, then subsides in October–March with the high uncertainty in the first discharge months (April–June) evident in the historical, observed, and future flows. The change signal is maintained in the results from the bias-corrected data with both the MC and QC correction methods, showing a negative discharge signal in the future. The MC bias correction method predicted lesser discharges in the low flows and higher discharges in the peak flows as compared with the QC method.

The FDC (Figure 7) from this model shows a decrease in the future discharge as compared with the historical with a more pronounced negative change signal than what the EC-EARTH predicted. For the high flows (0–10% exceedance probabilities), the UC climate simulations from this driving model predict a decline in the future flows and the corrected simulation shows a less significant change for this type of high flows relating to high rainfall events. The other segments of the FDC (10–100% exceedance probabilities) show a decrease and a negative change signal for the future flow with the UC data predicting a larger decrease in the future compared with the corrected climate simulations.

Overall, this model yields higher discharges in comparison with the other models. For example, the 25, 50, and 75% exceedance probabilities give discharges of 400, 40, and 10 m³/s, respectively. The discharge curve obtained from the historical RCM simulations as input is far away from the curve of the observed discharge, and the bias correction methods applied in this study tried to improve the RCM simulations. Both the bias correction methods maintained the negative change signal. The historical discharge as a result of correction using QC was better compared with the discharge from the MC correction method.

The annual cycle showed a decrease in the discharge for the future and a unimodal discharge cycle and emphasized that the corrected data using the QC method performed better than those using the MC method with the bias-corrected data performing better than the UC (Figure 8).

In the FDC of this model (Figure 9), the UC climate data show both an increase and a decrease of flows for the future. The flows from both the UC and corrected climate simulations produced a positive change signal for the high flows (0–10% exceedance probabilities) characterizing high watershed water yield from high rainfall events in the future. However, for flows in the wet to low-flow conditions (10–100% exceedance probabilities), this driving model predicted a huge decrease in the future discharge for the corrected climate simulations, bringing into question the future sustainability of the streamflows in the subregion in the long term and the intermediate base flow relaxation response of the catchment water yield. For 50% of the time, the flows from the QC and MC method corrected data indicated that a discharge of 5 m³/s will be equaled or exceeded in the future as compared with 15–20 m³/s in the historical period (Figure 9).

The annual cycle shows a slight increase in the peak discharges for the future as compared with the historical (Figure 10). The discharge for the historical period has a similar flow pattern with the observed discharge but the peak discharge is almost doubled in September. The month of the peak discharge is shifted from August in the observed discharge cycle to September in both the historical and future discharge cycles.

Figure 11 shows the FDC obtained when the results of the MIROC driving model were used as the input to the hydrological model. For the high flows (0–10% exceedance probabilities of flow) as it occurred in the other driving models, the flows as a result of using the MIROC model climate simulations predict an increase in the future flows which characterizes higher water yield in the catchment from high rainfall events. The flows in the other segment (10–100% exceedance probabilities of flow) rather predict a negative change signal (decrease of flows) in the future in very huge amounts. The corrected data also predict a negative impact signal on the hydrology of the catchment in the future. For example, the 50% exceedance probability of flow gives a discharge of 8 m³/s for the future and 20 m³/s in the historical for the corrected climate simulations (Figure 11). The results from the correction using the QC method were better than the MC method for the historical corrected simulations.

The annual discharge cycle (Figure 12) for both the corrected and UC data showed the same decrease of flows for the future for flows below 150 m³/s from January to the middle of August. Then for flows exceeding 300 m³/s which occurs around August/September, there is an increased discharge for the future as compared with the present. It could therefore be inferred that in the future, the dry periods become drier and the wet periods become wetter with more extreme cases happening. The month for the peak discharge is shifted from August to September as seen in the observed discharge curve. The future from this GCM driving model shows that from November to April, there is very little discharge and the rivers in the catchment virtually dry up. The flow begins to increase from July which is very late as compared with the observed and historical cycle, indicating a shortened discharge regime from 7 months to just 4, which might have dire consequences for the agriculture, hydropower, and water supply of the countries within the subregion.

The study demonstrates the effectiveness of using simple and less data demanding tools such as the lumped version of the HBV hydrological model and FDCs in quantifying the hydrological consequences of CC in the subregion after performing bias correction on the climate simulations. In this study, downscaled and corrected climate simulations of five GCM driving models were used in conjunction with a calibrated and validated HBV model to determine the consequences of CC on the hydrology of the Black Volta Basin. The bias correction of the various climate simulations used for the study was done using QC and MC methods. Although the study was completed in the year 2016, we believe it is still relevant and it demonstrates some interesting and revealing results for the current study area.

The results of this study show that all except one of the GCM driving models project a decrease in discharge for the future. The bias-adjusted (corrected) data confirming a decrease (negative impact) for the future flows. In the high flows (0–10% exceedance probabilities of the FDCs), there is a slight increase in the future flow characterizing a higher watershed water yield from high rainfall events in the future. The 10–40% exceedance probabilities of flow, representing wet conditions; 40–60% relating to mid-range flows; 60–90% meaning dry period conditions; and low flows (90–100%) all show a decrease in the future flows for four out of the five GCM driving models. Most concerning is the negative change signal (decreased flows) for the low flows in the future relating to the sustainability of streamflow in the long term. This could lead to the drying-up of the watershed in the long term.

This study has several strengths. First, we were able to bias correct the climate simulations using measured (observed) rainfall and temperature data as the ‘correct’ data. Second, we used two different correction methods in this study and their accompanying effects on the hydrology of the subregion. However, a smaller subset of (i.e. five) GCM driving models, two downscaling models, and one climate scenario (RCP 4.5) were used for the study, which might limit the overall results and conclusion of the study. Further research should be done with more GCM driving models and RCMs.

Our study slightly differs from the studies (Kunstmann & Jung 2005; Jung 2006; Jung & Kunstmann 2007; Jung et al. 2012; Roudier et al. 2014) in which the effects of CC on the hydrology of the Volta basin in Western Africa showed a very heterogeneous response of river runoff to changes in climate variables. Their work showed that there was no major change for the mean runoff regime of the future as compared with the current simulation. This discrepancy may partly result from differences in the climate models and the versions used for the various studies. However, our results are consistent with the results of the studies (Götzinger 2007; Giertz et al. 2010; McCartney et al. 2012; Kankam-Yeboah et al. 2013) that there is a significant decrease in the projected streamflow in the future for the West African subregion.

A particular strength of this study is the use of the daily temporal resolution of climate variables and streamflow datasets with a longer period (45 years) that provides a sufficient sample size for CC studies and impacts. Nevertheless, a major limitation of this study is the use of areal precipitation and areal climate variables which may be less precise to represent such a large catchment compared with previous studies that used distributed or semi-distributed climate datasets and hydrological models.

This study adds to the knowledge of CC impacts and also provides a simple, rapid, and powerful tool for assisting water stakeholders in identifying the flow regime components and formulating mitigation and management strategies. The information gathered may be useful in the development of new sustainable water-related projects and river basin management systems.

A number of questions remain unanswered, such as why some of the climate models give conflicting signal change results and as a result different impacts to hydrology and why most of the GCMs and RCMs are still not able to well capture the observed and historical climate. Therefore, additional research is needed to continually update the results with new models and scenarios to improve the acceptance of these results by society and stakeholders.

This study revealed that, except for the CanESM driving model which showed an increase in the future discharge, all the other GCMs (EC-EARTH, ECHAM, GFDL, and MIROC) projected a decrease in discharge for the future. The bias-adjusted data using QC and MC methods also confirmed a decrease (negative impact) for the future flows. The first segment of the FDCs, representing high flows (0–10% exceedance probabilities), shows a slight increase in the future flow characterizing a higher watershed water yield from high rainfall events in the future. The other segments of the curve, i.e. 10–40% exceedance probabilities of flow representing wet conditions; 40–60% relating to mid-range flows; 60–90% representing dry period conditions; and low flows (90–100%) all show a decrease in the future flows for four out of the five GCM driving models. Most worrying is the negative change signal (decreased flows) for the low flows (90–100% exceedance probabilities) in the future relating to the sustainability of streamflow in the long term. This could lead to the drying-up of the watershed in the long term.

It could therefore be concluded that CC will negatively affect and impact the hydrology of this subregion because there would be a decrease in the future discharge for the area. For flows occurring with 50% exceedance probabilities in the EC-EARTH driving model, a discharge of 18 and 28 m³/s is obtained for the future and historical periods, respectively.

The annual discharge cycles from the driving models showed a mix of unimodal and bimodal cycles. The MIROC model showed a delayed discharge cycle from August to September and a shortened discharge regime from 6 to 4 months. The 25, 50, and 75% exceedance probabilities produced discharges of 30, 6, and 2 m³/s, respectively, in the future for the MIROC model. The discharges using the ECHAM model were showing higher values in comparison with the other models.

In general, the decrease was more pronounced in the low flows (i.e. flows less than 100 m³/s). The results from the bias-corrected data looked better than the UC with the corrected data using QC performing better than the MC method.

In ameliorating the effects of CC on the hydrology and water resources of the region, proper adaptation and water resources management measures need to be put in place. It is also recommended for the region to possibly look for alternative sources of electricity because there would not be enough water in the reservoirs for hydropower production in the future. The negative effects of CC on the water resources are also a regional security threat because it may lead to loss of jobs, occupation, and farming harvest. This may subsequently destabilize the already vulnerable subregion and increase food insecurity and poverty. To mitigate food insecurity and poverty in the subregion, more irrigation dams need to be constructed to store the lower future water resource for farming activities.

We wish to state that this paper is born out of the doctoral dissertation of the first author and as such there can be overlap between the dissertation, which can be accessed at elib.uni-stuttgart.de, and this journal article.

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

Data cannot be made publicly available; readers should contact the corresponding author for details.