Projections and impact assessment of the local climate change conditions of the Black Volta Basin of Ghana based on the Statistical DownScaling Model

Global or large-scale Climate Models (GCMs) are presently employed in the development of scientific understanding and knowledge of climate change and variabilities in large-scale climate data (Dixon et al. 2016). Data acquired from GCMs provide in-depth comprehension of the present and future climate at a holistic scale (IPCC 2007, 2014). The >100-km resolution of GCM outputs renders them too coarse to be utilized in impact studies, decision-making processes, and adaptation planning at a local or regional scale (Meenu et al. 2013). Additionally, the limitations on the applicability and usefulness of GCMs in site-specific impact assessment studies are caused by the uncertainties and biases associated with GCMs that increase from global to regional to local scales (Lutz et al. 2016). Therefore, downscaling is required to increase the spatial resolution, reduce bias, and help accelerate climate forecasts for impact analysis and adaptation strategy. Presently, there are two types of downscaling techniques available to reduce biases and improve the spatial resolution of GCMs. These include statistical and dynamic downscaling techniques. The dynamic downscaling such as CORDEX (http://www.cordex.org/domains/region-5-africa/) contains local data which comprises topographic components that produce a high-resolution climate forecasting, for instance, 50 km for Africa (Dosio & Panitz 2016). One disadvantage of the dynamic downscaling process is its high computational resource requirements. Therefore, these models have inherent sensitivity and biases to the precinct of GCMs, and this limits their utilization in impact assessment and adaptation studies at the local scale (Wilby & Dawson 2007). Relatively, statistical downscaling models are simple, efficient, and fast which require low computational powers and expenses.

Statistical models are built to offer site-specific weather series that are equivalent to station data through the development of statistical relationships between large-scale climate variables also known as predictors and site-specific climate variables known as predictands. The advantages of the statistical downscaling techniques are that they are extensively employed in impact studies such as water and agriculture at the regional and local scale (Wilby & Dawson 2007; Tavakol-Davani et al. 2013). Generally, statistical models are categorized under three classes premised on the statistical approach; weather typing (Anandhi et al. 2011), weather generator (Semenov & Barrow 1997), and transfer function (Wilby & Dawson 2007).

The Statistical DownScaling Model (SDSM) (Wilby & Dawson 2007) is among the extensively applied statistical downscaling models that are built premised on a transfer function and stochastic weather generator (Hassan et al. 2013). Furthermore, the SDSM was found to be better than conventional weather generators in terms of performance (Mohorji et al. 2017) especially in capturing precipitation (Prcp), minimum (T_min), and maximum (T_min) temperature (Wilby & Dawson 2007; Hassan et al. 2013). Therefore, the SDSM is employed to simulate station-specific high-resolution climate projections for the Black Volta Basin (BVB). Most regions in Africa, especially in West Africa, are highly susceptible to changes in climate extremes such as weighty rainstorms, droughts, and frequent inundations and will be more severe considering future projections (Gebrechorkos et al. 2019a). To contemplate the observed susceptibility and changes of the BVB to variability (for instance precipitation variability) and changes in climate and extremes (Gebrechorkos et al. 2019a), undertaking a detailed impact study on the local scale is required to reduce or mitigate the effects in the forthcoming years via sustainable adaptation procedures. However, this category of data is scarce, and attaining site-specific climate projections using the SDSM needs high-resolution observed data for calibration and validation, as well as the ensemble of synthetic weather series employing daily GCM predictors (Wilby & Dawson 2007).

Observed weather ground station data in Ghana is generally limited due largely to issues such as limited spatial and temporal, accessibility (data distribution policies), and data quality. As a result, the dataset used was significantly influenced by the availability and consistency of the data collected. This, therefore, influenced the selection of the stations used in the study.

The application of the SDSM has been widely established throughout the world over the last decade for future climate change assessment at local scale. For instance, previous studies (Liu et al. 2011; Tryhorn & Degaetano 2011; Campozano et al. 2016) compared the performance of the SDSM to other models such as the Markov model, artificial neural networks, and least square support vector machine. Liu et al. (2011) found that the SDSM performed well in simulating the inter-site correlation features of the weather stations compared to the Markov model which simulated the multi-site prediction. Moreover, while assessing the extreme precipitation over the northern United States (Tryhorn & Degaetano 2011), it was found that the SDSM is capable of matching observed extreme climatology well compared to spatial disaggregation, and bias correction. Furthermore, Campozano et al. (2016) revealed that the SDSM is capable of simulating the median and variance of the depths of the monthly ground station data better than the artificial neural networks and least square support vector machine.

In Africa, a few studies (Wilby et al. 2014; Gulacha & Mulungu 2016; Matthew et al. 2017; Gebrechorkos et al. 2019a; Iwadra 2019) have successfully applied the SDSM with good performances. For example, Gebrechorkos et al. (2019a) found that the SDSM recorded good performance (>80, 79, and 96%) during validation for precipitation, minimum, and maximum temperature, respectively, in East Africa. Gulacha & Mulungu (2016) also found the performance of the SDSM ranging from 21–63% for precipitation and 60–98% for temperature in the Wami-Ruvu river basin in Tanzania. Again, the findings of Iwadra (2019) revealed the performance of the SDSM between 75–85% in the Aswa catchment of Uganda.

Currently, there is little or no study on the application of the SDSM in Ghana, especially in the BVB. Majority of the studies (Matthew et al. 2017; Osei et al. 2019; Awotwi et al. 2021; Sekyi-annan et al. 2021) conducted in Ghana applied the CMHyd model to bias-correct the hydro-climatic variables, especially under the Representative Concentration Pathway (RCP) scenarios. Studies on the A2 and B2 scenarios are still lacking in the BVB. Although these are old scenarios, their importance in future decision-making cannot be overlooked. It is worth noting that these studies coupled the CMHyd model with other hydrological and hydraulic tools such as the Soil Water Analysis Tool (SWAT), MIKE, HEC-HMS, etc., necessitates historical and future data as tangible inputs for the assessment of the impacts of climate change on hydrological resources at the basin scale. The BVB hosts one of Ghana's hydropower dams which provides 400 MW of power to the electricity grid in Ghana. Furthermore, the BVB extends over productive agricultural lands whose predominant activities are extensively rainfed and livestock farming. The past and future impacts of climate variabilities and changes on water availability in the BVB have been established in previous studies (Akpoti et al. 2016; Kabo-Bah et al. 2016).

Therefore, the main objective of this study is to calibrate and validate the analog technique as a method to downscale daily precipitation (Prcp), minimum (T_min), and maximum (T_max) temperature for specific stations of a data-scarce basin such as the BVB. This is critical to driving impact assessment models to evaluate the impact of future climate changes on sectors such as agriculture, hydrology, and urbanization in the basin which addresses the sustainable development goals 11 and 13 (UN 2023). In this study, future climate changes were downscaled under multiple climate scenarios based on three climatic periods 2011–2039, 2040–2069, and 2070–2099 as 2020s, 2050s, and 2080s, respectively. The outputs presented in the Innovative Trend Analysis (ITA) will assist in the identification of future climate trends in a detailed manner that could support the development of sustainable mitigation and adaptation strategies in the basin based on the climatic periods.

The A2 climate scenario describes a world of self-dependence and securing local identities. Thus, where countries operate independently, under this scenario, there is a slow convergence of the patterns of fertility across regions. This results in a rapidly high population. Also, there is an expansion in economies, per capita, and technologies. However, these expansions are regionally centered leaving them more fragmented than other narratives (IPCC 2007).

However, under the B2 climate scenario, the world tends to focus on local solutions to tackle environmental, economic, and social sustainability. Population under this scenario is expected to increase, however, at a lower rate compared to the A2 climate scenario. Similarly, there is an expected increase in economic development but at intermediate levels. This is expected to be fairly rapid, however, expansions in technologies are expected to be more heterogeneous by the end of the 21st century. Despite this scenario being social equity and environmental safety oriented, it focuses on local and regional levels more than the other narratives. This results in varying global warming under these scenarios.

For the Coupled Model Intercomparison Project 5 (CMIP5) under the RCPs, there are three different climate scenarios ranging from RCP 2.6 (the scenario where there is low emission as a result of active mitigation), RCPs 4.5 and 6.0 (the intermediate scenarios) as well as the RCP 8.5 (worst-case scenario/high ended emission scenario). Furthermore, the RCP scenarios have probable combinations of expected high population, energy intensity, and socio-economic development. The naming of the RCP scenarios follows their total radiative forcing by/or post 2100. For instance, under the RCP 2.6, the radiative forcing is expected to peak at approximately 3 W/m² in the mid-century before dropping to 2.6 W/m² by 2100, whereas the radiative forcing is expected to plateau without spiking at 4.5 W/m² by/post-2100 under the RCP 4.5 and rise to 8.5 W/m² by 2100 under RCP 8.5 (van Vuuren et al. 2011).

The ground station Prcp, T_min, and T_max data were obtained from the Ghana Meteorological Agency (GMet). The dataset used for all the selected stations spans from 1976 to 2016. These stations are located in the BVB (Figure 1). The dataset used was greatly influenced by the availability and consistency of the data collected. This, therefore, influenced the selection of the stations used in the study. For the GCM output, the study used 26 predictors (see Table 1) from the National Centers for Environmental Prediction (NCEP) and the National Center for Atmospheric Research (NCAR) reanalysis data of the Canadian Earth System Model version 2 (CanESM2) as well as the NCEP reanalysis data of the Hadley center Coupled Model version 3 (HadCM3). These reanalysis datasets span from 1961 to 2005 and 1961 to 2001 for NCEP-CanESM2 and NCEP-HadCM3, respectively, and were used for calibration and validation of the models. However, the CanESM2 and HadCM3 GCM data belong to 1961–2100 and 1961–2099, respectively. The CanESM2 and HadCM3 have a spatial resolution of about 2.81 and 2.5°, respectively, with uniform longitude and latitude. The CanESM2 and HadCM3 were used for the future projections downscaling from 2006 to 2100 and 2002 to 2099, respectively. The predictors from the CanESM2 were derived under the RCP 2.6, 4.5, and 8.5 scenarios whereas predictors from the HadCM3 were derived under the A2 and B2 scenarios. The six stations used for the study were located in a single GCM grid box (Box_90X_29Y and Box_128X_36Y for CanESM2 and HadCM3, respectively) for both models.

The next stage is the reproduction of daily ensembles for the current period using a well-calibrated model. These models are afterward subjected to the evaluation of their performance using the Taylor diagram and several dimensionless and dimensioned statistics as well as goodness-of-fit. Future climate is subsequently projected based on the selected predictors obtained from the GCM outputs. Finally, the Innovative-Sen trend plots were used to analyze the trend patterns to identify the partial and overall trends in the future climate. However, the delta statistics (Wilby & Dawson 2007) were used to assess the magnitude of changes found between future and baseline periods. It is worth noting that the model calibration and scenario generator SDSM functions were used to calibrate and generate future scenarios, respectively.

For the evaluation of model performance, the Taylor diagram (Taylor 2001) was used to compare the ensemble mean with individual ensemble members during calibration and validation. The Taylor diagram was used because of its ability to graphically measure the standard deviations, root-mean-square difference, and correlation coefficient of the models and observed data.

The overbars show the means of the daily values for the whole analyzed period. The scaling coefficient c should preferably be set to 2 (Legates & Mccabe 2013; Willmott et al. 2015). In estimating the index of agreement with baseline adjustment (⁠⁠), the mean of the daily values (⁠⁠) is replaced by the mean of monthly values (⁠⁠.

On model performance evaluation, using graphical techniques only is not highly recommended. Therefore, in addition to graphical methods, both dimensionless and dimensioned measures are highly recommended (Legates & McCabe 1999; Willmott et al. 2015). Table 2 presents dimensioned statistics during calibration and validation results for both models.

For the dimensioned statistics, the values of MAE during calibration ranged from about 0.342–0.574 mm, 0.007–0.093 °C, and 0.011–0.285 °C for Prcp, T_max, and T_min, respectively (Table 2). It is therefore obvious that the errors of the model during calibration for T_min are slightly higher than T_max. To compare the two models (HadCM3 and CanESM2), both simulated daily Prcp with similar errors during calibration. However, CanESM2 generally recorded higher errors in predicting daily Prcp during calibration. During validation, the values of MAE ranged from 0.351–1.060 mm, 0.089–0.889 °C, and 0.069–0.295 °C for Prcp, T_max, and T_min, respectively. This reveals slightly higher model errors during validation for both models especially in simulating T_max. However, this is consistent with the findings of Phuong et al. (2020). There were fairly similar model errors for both models (HadCM3 and CanESM2) in simulating the daily Prcp and T_max during validation. However, CanESM2 recorded more significant model errors than HadCM3 in simulating daily T_min during validation. The values of MBE ranging from −0.342 to −0.574 mm, −0.059 to −0.001 °C, and −0.128 to −0.0001 °C for Prcp, T_max, and T_min respectively show a slight underestimation of predictands during calibration. To compare both models, CanESM2 was found to significantly underestimate daily Prcp and T_max compared to HadCM3 (Table 2) during calibration. MBE values ranged from −0.627 to −0.042 mm, −0.049 to 0.187 °C, and −0.274 to −0.043 °C for Prcp, T_max, and T_min, respectively. Similarly, CanESM2 underestimated daily Prcp, T_max, and T_min compared to HadCM3 during validation (see Table 2). However, the dimensioned statistical results on validation were generally satisfactory since they fell within an acceptable range on the daily time scale.

Moreover, the goodness-of-fit results from R² revealed better performance during calibration and validation. For instance, the values for R² ranged from 0.997–0.999, 0.950–0.999, and 0.9713–0.996 for T_max, T_min, and Prcp, respectively, during calibration. Surprisingly, the R² values for precipitation were higher during calibration (i.e. between 0.971–0.996). Similarly, other studies (Gebrechorkos et al. 2019b) revealed higher performance of the SDSM based on R between 0.79–0.98 in basins of East Africa. The performance improvement observed in this study may be attributed to the long period of calibration (ranging from 25–29 years), which is noted to impact the model's performance. The longer period of calibration is found to generate probability distribution functions in the process of statistical downscaling (Kim et al. 2000). However, the values for R² dropped during validation as they ranged from 0.802–0.999, 0.955–0.997, and 0.646–0.957 for T_max, T_min, and Prcp, respectively. The performance of HadCM3 in simulating the daily Prcp during validation was higher than CanESM2 (see Table 2). However, the performance of both models was similar in simulating daily T_max and T_min during validation. Although the determinants of the goodness-of-fit dropped generally during validation, the performance of both models was better compared to previous studies (Gulacha & Mulungu 2016; Tarek et al. 2016) in Africa. Furthermore, the length of data for calibration may have some bearing on the performance of the model. This helps the SDSM to generate probability distribution functions in the process of downscaling (Kim et al. 2000). The performance of both models revealed by the goodness-of-fit results during calibration was very high and therefore increased the confidence level of the future projections.

To some magnitude, the findings of this study reveal numerous uncertainties in projecting future daily Prcp. Phuong et al. (2020) noted that the projected climate uncertainties are mostly attributed to internal climate variabilities, downscaling tools and methods, and the data used (which includes predictands and predictor variables used). The study employed CanESM2 and HadCM3 predictors which represent manifold emission scenarios. Therefore, other downscaling techniques such as LARS-WG, Bias correction, and Machine Learning approaches can be applied along with the SDSM (Etemadi et al. 2014). Phuong et al. (2020) also recommended the use of more than one potential predictor set for individual climate data at each station to reach a plausible range of climate variation in future periods.

Moreover, there is an expected high warming tendency in the BVB as T_max is projected to record a high magnitude of warming at all the selected stations (Figure 7). Figure 7 reveals that there is an expected minimal variation in T_max in the 2020s under all the scenarios. However, the T_max is expected to vary significantly in the magnitude of warming in the 2050s and 2080s under all the scenarios. For instance, T_max is projected to increase by 0.008–0.52 °C in the 2020s compared to the baseline period. However, T_max has a high tendency to generally increase under both models by 0.007–1.39 °C in the 2050s and 0.22–2.44 °C in 2080s compared to the baseline period. However, the highest T_max is expected to occur at Wa under the A2 scenario (see Figure 7). Sunyani is projected to be the coldest station compared to the other stations (where the change in T_max is <0.8 °C). Similarly, T_min is generally expected to generally increase by 0.009–0.47, 0.006–0.65, and 0.002–1.83 °C as projected by both models across the periods (Figure 7). Generally, the projections of HadCM3 for T_max and T_min under the A2 and B2 scenarios were found to be higher than that of the CanESM. The findings of this study are in line with a previous study that found that temperatures are projected to increase in the BVB (Siabi et al. 2021).

Aside from estimating the delta statistics for trend analysis purposes, the study applied the ITA technique to the selected stations. For interpretation of the ITA plot, it is necessary to add the ±10 error line and divide the plot into three segments that are low, medium, and high. The purpose of these error lines is to ensure a clear understanding of the spacing of the data points from the trendless line without any statistical connotations. Thus, the study also relates to the identification of partial and overall trends between the 2020s, 2050s, and 2080s and the baseline period under different scenarios. The Innovative-Sen trend plots were applied to ascertain whether the annual climate is expected to increase or decrease based on the location of the climate variable falling on the plot.

The decline in Prcp revealed by the findings of the study agrees with the findings of Yira et al. (2016) where Prcp is expected to drop by 9–12%. The findings of Gulacha & Mulungu (2016) also revealed a change in Prcp by −44 to 648% and −37 to 346% under the A2 and B2 scenarios, respectively. Moreover, Asante & Amuakwa-Mensah (2015) revealed that climate change together with detrimental anthropogenic activities concerning land use may worsen desertification in northern Ghana. Population growth which triggers expansion of settlements, agriculture, and over-exploitation of natural resources in an unsustainable manner in a desertification susceptible zone of Ghana may be a cause. However, this is worsened by the decline in Prcp in northern Ghana. This reveals a high probability of desertification.

The mixed trend of the expected Prcp under all the scenarios agrees with the findings of Kebede et al. (2013) where Prcp in the Baro-Akobo basin of Ethiopia is expected to change by −2 to 21%. Therefore, Yira et al. (2016) noted that the negative and positive Prcp trends reemphasize the discourse on the importance of the trend rather than the magnitude of change in projected Prcp over West Africa.

Generally, T_max exhibits a significant increasing trend under all the scenarios as most of the scatter points fell in the upper triangular area of the plot (Figure 9). For Babile, there is an expected low to medium increasing trend in T_max especially in the 2080s under the RCP 8.5. This is denoted by purple square points falling at the upper diagonal. However, this increase is expected to be gradual since the point fell within the lower to medium part of the upper diagonal (see Figure 9). Similarly, stations such as Bui, Wenchi, and Sunyani are expected to observe a low to medium increase in the 2080s especially under RCP 4.5 and the worst-case scenario (RCP 8.5). Also, Wa and Bole are expected to observe a gradual increase in T_max in the 2080s, however, this is expected to occur under the A2 scenario. This is signified by the blue square points scattering in the lower to medium part of the upper diagonal (see Figure 9). The projected increasing trend in T_max is in line with the findings of Kaboré et al. (2015) and Badou et al. (2018). For instance, Kabore found that T_max is expected to increase by 1.8 and 3.0 °C under the RCP 4.5 and RCP 8.5, respectively, in the Massili basin of Burkina Faso. Similarly, the study of Oyerindé (2016) revealed that temperature is expected to increase between 5 and 10% under RCP 4.5 and 5 and 20% under RCP 8.5 in the Niger river basin by the end of the 21st century. Moreover, Kebede recorded an increase in temperature under the B1 scenario relative to A1B for 50 and 40% of the stations for T_max and T_min, respectively.

Although the findings of these studies (Kebede et al. 2013; Kaboré et al. 2015; Oyerindé 2016; Badou et al. 2018) are higher than the current study, this could be largely due to the relatively high temperatures observed in Niger and Burkina Faso compared to Ghana. Conversely, stations such as Wa, Bui, Wenchi, and Sunyani are expected to record a gradual decrease in T_max in the 2020s under the B2 scenario. This is signified by orange round points falling in the low to medium segment below the 45° line (Figure 9). However, the decreasing trend in T_max is expected to be high at Wa in the 2020s under the B2 scenario (Figure 9). Also, Bole is expected to record a gradual decrease in T_max in the 2020s, however, this is expected to occur under the A2 scenario.

T_min is projected to generally follow the gradually increasing trend of T_max into the future under all the scenarios. For example, T_min shows a gradual increase, especially in the 2080s under the RCP 8.5 scenario at Babile, Wa, and Bole (see Figure 9). This is shown by the purple square point scattering in the lower to medium part above the 45° line. Similarly, T_min is expected to increase gradually at Bui and Sunyani in the 2080s under the RCP 8.5. For some reason, T_min is expected to decrease gradually in all three periods under all scenarios at Wenchi. This is seen as most of the scatter points fell at the lower triangular portion of the plot (see Figure 9). For the assessment of partial-trend, T_max, and T_min show the same partial-trend patterns. These partial trends are particularly likely to be more tangible when their respective values increase, indicating an obvious occurrence of hot weather events in future periods. However, the decreasing trends in T_min at some stations show the discontinuation of warming by the end of the 21st century. This is in line with the findings of the IPCC (2014). This validates the findings of Kebede et al. (2013) after downscaling T_min and T_max under the A1B scenario. T_min was found to decrease in the future (Kebede et al. 2013), however, the decrease in T_min is expected to be drastic compared to the current study.

In the face of higher temperatures, an outbreak of diseases is expected to increase. Previous studies have revealed the outbreak of diseases such as measles, cholera, diarrhea, guinea worm, and cerebrospinal meningitis among others as a result of the increase in temperature and decrease in Prcp in Ghana (Asante & Amuakwa-Mensah 2015). Again, an increase in temperatures coupled with poor drainage systems may increase vector-borne diseases such as malaria Haines et al. (2006) found that climate change may affect the spread of mortality and morbidity through physical impacts of exposure to high and low temperatures.

Moreover, crop yield in the basin is expected to be affected due to drought. Kusakari et al. (2014) found that the Upper West region located in the northern part of the BVB of Ghana is vulnerable to climate variability and extreme conditions such as floods and droughts severe temperatures. This affects agriculture which serves as the major livelihood (GSS 2021), and renders it defenseless in the face of harsh climate conditions. Studies (Dayour et al. 2014; Gordon et al. 2019) revealed that most agricultural activities in the north of Ghana rely tangibly on the rainfed system, making them susceptible to climate-related extremes. For instance, rice is a major crop produced in northern Ghana that runs on irrigation schemes and the rainfed uplands and lowlands system. Therefore, under extreme conditions, rice production is likely to drop. Knox et al. (2011) found that there is a projected variation of about −8% in rice production. Again, wheat production is expected to drop by about 99% over West Africa in the 2080s. However, maize production is expected to drop by 1–7%. Also, Boatemaa et al. (2022) revealed that crop irrigation requirements for tomatoes at Bole are expected to increase by 21 and 27% under RCPs 4.5 and 8.5, respectively, in the 2080s. Perennial water scarcity is known to be predominant in the Savannah zone where the BVB lies. This poses a threat to dry-season farming, mostly rainfed in the basin. For this reason, there is an expected increase in food prices and seasonal migration of households to southern Ghana in the dry season in quest of fertile lands and mining sites to feed family members in the future.

Moreover, Bessah et al. (2022) showed that the stations in the transitional zone (Wenchi and Sunyani) aligned to one of the three major climatic zones characterized by similar factors while delineating the climatic zones in Ghana. Therefore, a future shift in temperatures, rainfall seasonality, or monthly patterns may trigger a potential reclassification of the agroecological zones in the BVB. This will help in formulating adequate policies to manage climate-related interventions such as planting for food and jobs. Future water requirements for the production of food and energy as well as domestic sustenance are expected to surge by 34% by 2030 (Opoku et al. 2022).

Flooding events are expected to occur in the face of increased precipitation events. This calls for mitigation and adaptation strategies for floods, especially in the 2050s and 2080s. The report of OCHA (2007a) recorded the negative impacts of a catastrophic flood that hit West Africa in 2007. This led to the loss of lives, severe food shortage, and polluted water bodies in the BVB. About 56 people lost their lives and 332,600 people were affected in the northern, Upper West, and East region of Ghana (OCHA 2007b). Sawai et al. (2014) noted that, despite the BVB being located downstream in the semi-arid zone, floods and droughts are the main challenges. For instance, another devastating flood that occurred in August 2021 after a 12-hour downpour in the Upper West region left some 1,605 people in the Jirapa district, Lawra municipality, and the Nadowli-Kaleo district greatly affected (Mordey 2021). About 722 farms were lost as a result. Again, approximately 336 people were displaced and 155 houses were destroyed. Also, farmlands, communities, and irrigation sites were flooded along the Nadowli-Tangasie highway. The main bridge on the Wa-Hale Road that links the Jirapa, Nandom municipalities, Nadow-Kaleo districts to Wa was washed away. This calls for an urgent climate adaptation and mitigation actions to abate the adverse effects of climate change to make cities in the BVB safer, sustainable, inclusive and resilient which addresses the sustainable development goals 11 and 13.

Finally, in the face of a drop in precipitation events, water levels in the basin are expected to fall, adversely affecting the Bui dam's hydropower operations. In 2014, the dam recorded a drop in water level by about 5 m below the maximum reservoir level of 183 m above sea level at an elevation of about 177.9 m above sea level (Bui Power Authority 2014). Therefore, the Bui Power Authority must plan strategically to meet extreme climate changes so that the dam can operate above the projected minimum of 172.09 m above sea level under varying climates by the end of the 21st century.

The study successfully employed the SDSM tool to downscale daily Prcp, T_max, and T_min and generated future climate projections using HadCM3 and CanESM2 GCM predictors over the BVB of Ghana. The performance of the model was evaluated based on Taylor diagrams, dimensionless, dimensioned, and goodness-of-fit statistics. The calibrated models can be utilized for future scenario development because the statistics of the validated models were within the permitted range. The performance of the SDSM was found to be better in downscaling daily T_max and T_min compared to Prcp as shown by the various statistics used.

The trend analysis of projected future climate changes by delta statistics conforms with the results of the ITA. T_max was projected to increase in the BVB at a higher degree compared to T_min. T_max and Tmin are expected to increase by 0.22–2.44 and 0.002–1.83 °C in 2080s, respectively, which are relatively higher compared to the values of the other scenarios as revealed by the delta statistics and the relative location of the scatter points on the ITA plots. Concerning daily Prcp, there was a higher magnitude of variability in the downscaled outputs compared to T_max and T_min. The future Prcp projected by HadCM3 and CanESM2 indicated a decreasing trend as revealed by the delta statistics and ITA plots. Both models projected near- to far-future decreases by 2.05–23.89, 5.41–46.35, and 5.84–35.33% in the 2020s, 2050s, and 2080s. In summary, the findings of this study infer that the BVB has a high tendency of becoming drier and hotter in future periods, which may lead to potential challenges relating to water, food, and energy consumption. The findings of the study are therefore expected to inform policies in Ghana, especially related to water, food, and energy in the basin, and are informative for pragmatic climate adaptation and mitigation strategies. This signifies there must be an urgent diversification of energy sources by policy-makers and stakeholders to develop a more resilient energy sector to support energy demands especially during severe water drops which leads to variabilities in hydropower generation. Therefore, energy providers as well as the government of Ghana must consider investing in renewable energy sources such as thermal power plants and solar energy to supplement electricity generation. Moreover, energy sector investors must critically consider the effects of rainfall and temperature on potential energy generation. Beyond this, the government and policy-makers must foster the ‘Green Ghana’ agenda to support the decentralization of the prime goal of climate resilience to the grassroot level. Again, policy-makers and stakeholders must consider formulating policies that aim at mitigating the negative impacts of floods and droughts such as the utilization of rainwater harvesting systems as a potential for drought and flood mitigation, the design of infrastructure and production schemes to cater for higher risks of damages due to drought or floods, the development and implementation of early warning systems as well as the effective and efficient management of various water and agricultural resource projects in the BVB. Such climate actions improve the interdependency between people and their natural environment. The findings of the study are expected to influence evidence-based decision-making by stakeholders to directly relate climate actions to pressing needs based on scenario modeling. However, the study was limited to CMIP3 and CMIP5, future studies should consider socio-economic parameters that influence climate change in the BVB. Therefore, detailed future analysis is required to explore the new CMIP6 data under the shared socio-economic pathway scenarios.

This research received no funding.

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

The authors declare there is no conflict.