Impacts of hydro-climatic trends and upstream water management on hydropower generation at the Bagré dam

Global hydro-climatic trends are ambiguous, challenging the management of water resources. This challenge is addressed in the current study by investigating the impacts of hydro-climatic trends and upstream water management on hydropower generation at the Bagré dam. Modi ﬁ ed Mann – Kendall trend test, Standard Normal Homogeneity Test and Pettit test were applied to some selected hydro-climatic variables for the trend and the change year detection, whereas the relationship between upstream dam management, hydro-climatic variables and hydropower were assessed through the Spearman correlation. The results revealed an annual positive trend for all hydro-climatic variables except for water level, lake evaporation and out ﬂ ow. The break years observed in hydropower generation (2002) and in ﬂ ow (2006) were mainly due to the construction of the Ziga dam in 2000 and its management change in 2005, respectively. The study also showed that hydropower generation declines each May ( (cid:1) 30.36 MWh) and June ( (cid:1) 16.82 MWh) due to the signi ﬁ cant increase in irrigation withdrawals (1.94 hm 3 in May and 0.67 hm 3 in June). The results of this study highlighted the non-linearity in the relationship between hydropower generation and hydro-climatic variables as none of the correlation coef ﬁ cients (apart from turbine) are very strong ( . 0.8). As many human activities occurred in the basin, further research should be focused on the use of semi-distributed models to assess the impacts of water-use and land-use change on hydropower generation.


INTRODUCTION
Climate change is one of the major threats faced by many regions across the world (Khalid et al. 2017).The Sixth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) published in 2021 highlighted an increase in global temperature of about +1.2 °C in 2020 compared to the period of 1850-1900, and this is expected to increase by 1.4-4.4°C in 2100 under the lowest and highest Green House Gases emissions scenarios (SSP1-1.9 and SSP5-8.5),respectively (IPCC 2021).The rate of the increase depends on the location.In the Eastern Himalayas, for instance, the mean temperature increased by 0.02 °C This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY 4.0), which permits copying, adaptation and redistribution, provided the original work is properly cited (http://creativecommons.org/licenses/by/4.0/).per year while in Nepal the rate of increase was 0.04 °C per year from 1977 to 2000 (Khalid et al. 2017).The monsoon precipitation also experienced a decrease from the 1950s to the 1980s and more extreme events since the 1950s (DSB 2011;IPCC 2021).For instance, some droughts in East and West Africa are attributed to the increase of sea surface temperatures (SSTs) in the Indian and the Atlantic Oceans (DSB 2011).Whereas, Western Europe experienced in 2021 a huge flooding which caused 179 deaths in Germany and 36 in Belgium (WMO 2021).These changes in the climate patterns, mostly attributed to human-induced activities (IPCC 2021), have an impact on river discharge and water levels of reservoirs.The seasonality of the global river discharge (increase of high flows and decrease of low flows) is expected to increase by one-third during the period 2070-2100 compared to 1971-2000 (Michelle et al. 2013).In the Midwestern United States, the streamflow is predicted to decrease by 41-61% for the period 2051-2095 relative to 1978-2009 (Chien et al. 2013).In China, a change of À3.6 to þ14.8% in annual rainfall will induce a change in annual discharge of À29.8 to þ16.0% (Birkinshaw et al. 2017).In Africa, an increase in rainfall of 5-25% by 2080 within the Lake Victoria River basin will lead to an increase in river discharge of 5-267% while a variation of À2 to 5% in the annual rainfall will lead to a change of À20 to 10% in river discharge in the Niger River basin (Stanzel et al. 2018;Olaka et al. 2019).These changes in river discharges impact water levels of reservoirs used for many activities, including hydropower generation (Boadi & Owusu 2017;Dias et al. 2018;Mousavi et al. 2018;Sridharan et al. 2019;Liu et al. 2020).
Many African countries rely on hydropower dams to generate electricity for their populations.According to Okudzeto et al. (2014), the reliance of societies on climate sensitive sectors such as hydropower has become a challenge to sustainable development as a result of climate variability and change impacts on hydrological processes.In semi-arid areas of West Africa, changes in rainfall are a concern because of their important role in irrigated agriculture, domestic and industrial water supply, reservoir operation and hydropower generation (Ibrahim et al. 2016).According to Obahoundje et al. (2018), the operation of West Africa's hydropower dams relies strongly on the hydro-climatic condition of their upstream catchments.In addition, hydropower generation could also be affected by other factors such as population growth, land-use land cover, water abstraction and socio-economic development (Boretti & Rosa 2019;Alsaleh et al. 2021;Obahoundje & Diedhiou 2022).Therefore, reliable information regarding hydro-climatic trends as well as upstream catchment water management is important to adapt to projected future climate conditions (Sankarasubramanian & Vogel 2003;Oguntunde & Abiodun 2012).Besides, Frumhoff et al. (2015) showed that there is a relationship between El Niño-Southern Oscillation (ENSO) and hydropower generation.As ENSO is a major driver of climate variability, there is a need to investigate the prospect of ENSO-driven variability in hydropower generation at dams located in affected river basins (Jia et al. 2017).
Burkina Faso is not exempted from impacts of climate variability and change (MEFR 2015).The country is characterized by high rainfall variability resulting in strong fluctuations of river discharge, negatively impacting water resources availability (Karambiri 2017).A study by Ibrahim (2012) analysed rainy season characteristics  in Burkina Faso and found an increased frequency of low rainfall (between 0.1 and 5 mm/day) and heavy maximum daily rainfall, and also projected an increase of the length of the mean dry spell as well as a late cessation of the rainy season (2021-2050).Tazen et al. (2012) observed that the Nakambé River Basin (NRB) is under anthropogenic pressure as several small dams were built to supply water for domestic, irrigation, breeding and fishery needs.Changes in climate and land management can result in higher frequencies of droughts and floods, leading to food insecurity, diseases, population migration and loss of natural resources (Tazen et al. 2012;Annys et al. 2020).Thiombiano (2011) assessed the variability of rainfall and temperature in the NRB from 1940 to 2008.However, there is a lack of scientific evidence on how hydropower generation is related to past hydro-climatic trends and upstream water management in the basin.This study tries to answer the following question: are the trends and changes observed in the natural climate system also detectable in the hydrological system and the reservoir system of Bagré?The objective is to assess the impact of the climate system trend and upstream reservoirs management on the hydrological and reservoir systems in the NRB.

Study area
The NRB is located in the headwaters of the White Volta Basin in Burkina Faso (Karambiri et al. 2011).The basin is located between longitudes 2 °10 and 0 °30 West and latitudes 11 °11 0 and 14 °10 North (Figure 1).Its area is about 32,623 km 2 at the Niaogho station and is partly located in the Sahelian and the north Sudanian climate zones.The basin experiences a unimodal rainfall regime with the rainy season from June-July to September-October.The mean annual rainfall varies between 500 and 800 mm from north to south.The Water Resources General Directorate highlighted the presence of more than 400 reservoirs in the NRB (DGRE 2010).The most important reservoirs are Bagré, Ziga, Tougou, Dourou, Loumbila, Bam and Dem (Figure 1).
The Bagré dam was built in 1992 in the downstream part of the NRB.It has a water storage capacity of 1.7 billion m 3 , a water surface of 255 km 2 , a length of 80 km and a width of 3-4 km (Zoungrana 2007).It is a multipurpose dam whose main purpose is to generate 44 million kWh of hydroelectricity per year, corresponding to 20% of the national electricity demand (Kabore & Bazin 2014).The installed hydropower plant (HPP) has a capacity of 16 MW (two units that operate simultaneously producing 8 MW each during peak periods).The dam is also used for fishery and irrigation water supply of around 6,000 ha of rice cultivation (Zoungrana 2007).

Data collection, quality control and pre-processing
The datasets used in this study are both climate (rainfall, minimum and maximum temperature, the Oceanic Niño Index (ONI)) and hydrological data (inflow, outflow, irrigation, turbined flow, evaporation, water level, Ziga volume and hydropower generation).Table 1 presents the spatial and temporal resolution of the data.
Two sources of rainfall data were used.These include observed and satellite-based data.Observed rainfall and temperature from two synoptic stations (Ouagadougou and Ouahigouya) were provided by the National Meteorological Agency (ANAM), whereas the satellite climate products were from the Climate Hazards Group Infrared Precipitation with Station (CHIRPS) and W5E5.CHIRPS is a quasi-global rainfall dataset comprising in situ station data and satellite imagery to form a gridded precipitation time series at 0.05°horizontal resolution (Funk et al. 2015;Ogega et al. 2021).W5E5 is a merged dataset.Version 2.0 of the W5E5 dataset combines WFDE5 data over land with ERA5 data over the ocean and covers the entire globe at 0.5°horizontal and daily temporal resolution from 1979 to 2019 (Hersbach et al. 2020;Lange et al. 2021).The satellite rainfall data were used to supplement the limited stations' data in terms of spatial coverage in the basin.All grid cells are located upstream of the Bagré dam (Figure 1), as inflows and water levels are strongly influenced by rainfall volumes upstream of the dam (Khan & Short 2001).
The ONI was taken from Golden Gate Weather Services (GGWS) at https://ggweather.com/enso/oni.htm.It presents the running 3-month mean SST anomaly for the Niño 3.4 region (5 °N-5 °S, 120°-170 °W).All the other data were collected from the General Directorate of Water Resources (DGRE) and the National Electricity Company (SONABEL) which is in charge of the Bagré dam operation.
Data quality control comprised the identification of missing data in satellite and climate station time series and their comparison in the period 1985-2015.No missing data were found in the time series of the two stations.As suggested by Okafor et al. (2017), visual inspection has been used to detect outliers and to ensure internal consistency.
All data have been aggregated at monthly and annual scales.Observational rainfall data were used to validate the satellite rainfall.For this purpose, the CHIRPS and the W5E5 rainfall data were compared to the station's rainfall at the monthly scale using statistical indicators such as Pearson correlation coefficient (R), Root Mean Square Error (RMSE), Nash-Sutcliffe Efficiency (NSE) and Percentage bias (PBias).The satellite product that showed the best performance (CHIRPS) was used to investigate the mean basin rainfall temporal trend, whereas the stations (Ouagadougou and Ouahigouya) were used to investigate the trend at different locations in the NRB.

Change year detection
Breakpoint detection in hydro-climatic time series helps to identify regime shifts (Badou et al. 2017).It aims at detecting abrupt changes, called breakpoints, in the distribution of a signal (Bock et al. 2020).Many methods exist for breakpoint detection in climate variability and change studies (Caussinus & Mestre 2004;Mestre et al. 2013).The Pettitt's test (Pettitt 1979) considers a sequence of random variables X1, X2, …, XT, which is said to have a change point at τ if Xt for t ¼ 1,…, τ have a common distribution function F1(x) and Xt for t ¼ τ þ 1 ,…, T has a common distribution function F2(x), with F1(x) and F2(x) different, but continuous (Rybski & Neumann 2011).In this study, the Pettitt's test and the Standard Normal Homogeneity Test (SNHT) were applied to the hydro-climatic variables (Khaliqa & Ouarda 2007;Gulakhmadov et al. 2020).

Trend analysis
Several statistical methods exist for detecting trends in climate variables.These are mainly composed of innovative trend analysis approach (Sen 2012), Cumulative Sum (CUSUM), Mann-Kendall (MK), Bayesian analysis, etc. (Sonali & Kumar 2013).Commonly used is MK (Sonali & Kumar 2013).In this study, the trends of hydro-climatic variables and the hydropower generation were analysed using the Modified MK trend test, a non-parametric trend test used to identify monotonic trends present in time series data (Mann 1945;Sen 1968;Kendall 1975).This statistical method requires no specific distribution of the data (Kendall 1975).The method is robust against outliers (Sanogo et al. 2015;Abungba et al. 2020) and can cope with missing values and values below the detection limit (Gavrilov et al. 2016).It measures the correlation of a variable with time and gives information on the direction, magnitude and significance of observed trends (Koudahe et al. 2017;Gulakhmadov et al. 2020).More information on the Modified MK test is available in the literature (Hamed & Rao 1998;Hamed 2008;Tabari et al. 2011;Gavrilov et al. 2016;Pingale et al. 2016;Abungba et al. 2020;Gulakhmadov et al. 2020).First, all time series were checked for autocorrelation.For cases of auto-correlated data, time series were pre-whitened before the trend computation.All statistical analyses were performed with the R software packages modifiedmk, hydroGOF, correlation and trend.The trend analysis was performed for annual and monthly time series of rainfall, mean temperature (minimum and maximum), inflow, outflow, irrigation, turbined flow, evaporation, water level, Ziga volume and hydropower generation.Results with a p-value 0.05 indicate a significant trend while a p-value .0.05 indicates an insignificant trend.

Correlation between hydro-climatic variables, ENSO and hydropower generation
Variability in hydro-climatic variables and upstream dam management influences downstream hydrology and hydropower generation (Beilfuss 2012;Annys et al. 2020).To assess the relationship between hydro-climatic variables, upstream dam management and hydropower generation, Spearman correlation tests were performed.The Spearman correlation test is a non-parametric test based on the rank and it measures the strength of the monotonic relationship between paired variables (Badou et al. 2017).The annual hydropower generation was correlated to annual rainfall sums, annual inflow, annual outflow, annual irrigation, annual mean water level, annual lake evaporation and annual Ziga volume over the period 1993-2012 and 2001-2012 (corresponding to the availability of hydropower generation and Ziga volume data).A 3-month mean basin rainfall, inflow, mean water level, lake evaporation and hydropower standardized anomaly indices were computed and correlated to the 3 months' ENSO index.Spearman rho results are interpreted as follows: 0-0.20 is negligible, 0.21-0.40 is weak, 0.41-0.60 is moderate, 0.61-0.80 is strong and 0.81-1.00 is considered a very strong correlation (Prion & Haerling 2014).

Satellite rainfall data validation
At the mean monthly scale over the period 1985-2015, the statistics showed a good correlation between the satellite products and the observed rainfall at the two stations (Table 2).The bias is not very significant, considering ranges from À20 to þ20% as an acceptable performance for a satellite product (Cohen et al. 2012;Diem et al. 2014).As the CHIRPS has a lower bias (1 and À1.9) compared to the W5E5 (3.8 and À8.4), it was chosen to analyse the mean basin rainfall trend.

Trend analysis and change of annual hydro-climatic variables and hydropower generation
The results of the Modified MK trend test are presented in Table 3.All variables showed an increasing trend over their respective considered periods.Rainfall, temperature, inflow, turbined flow, irrigation, Ziga volume and hydropower generation show significant increasing trends at the 95% confidence level, whereas the increasing trends of lake evaporation, outflow and water level were not significant.The annual slopes of rainfall were 5.27 mm over the basin, 6.17 mm at the Ouagadougou station, 8.77 mm at the Ouahigouya station, 68.08 hm 3 for inflow, 17.99 hm 3 for turbined flow, 12.54 hm 3 for irrigation, 2.15 hm 3 for Ziga volume and 2,168 MWh for hydropower generation.The results also show that minimum and maximum temperatures increased significantly by 0.04 °C/year at the Ouagadougou station, whereas the Ouahigouya station recorded a significant annual increasing slope of 0.02 °C.Besides, the results of SNHT showed that all variables experienced break years except mean basin rainfall, inflow and water level, whereas the Pettitt's test showed no break for only water level and maximum temperature at the Ouahigouya station.

Trend analysis of monthly hydro-climatic variables and hydropower generation
The Modified MK trend test was performed for the monthly times series of all variables.Tables 4 and 5 summarize the trend direction, the significance and the slope of each month over the variables and considered periods.Mean basin rainfall increased from May to October over the period 1985-2015.Only the month of July increases significantly at 99% (Table 4).At the station level, rainfall in Ouagadougou increased significantly in July and September (rainy season), whereas in Ouahigouya a significant increase in February (dry season) is observed.This significant increase of rainfall in the dry season has a negligible annual slope (Table 4) and is therefore not relevant in terms of water balance.The minimum temperature increased in all months at the two stations, whereas the maximum temperature increased differently.In Ouagadougou, the maximum temperature increased for all months, whereas in Ouahigouya, it decreased from August to October, but not significantly.
For lake evaporation, significantly increasing trends were found for September and December, whereas for March, August and October non-significant increasing trends were found.Decreasing lake evaporation trends were found for January, February, April to July and November, but these were not significant.Meanwhile, for inflow, the months of February to November show increasing trends between 1993 and 2018.However, significant trends are only noticed in May, July, August and September.During the same period, decreasing trends of mean lake water levels were found from March to July and increasing trends from August to February.Significant negative trends are observed in May and June while significant positive trends were perceived in September and October.Outflow increased from August to October, with a significant increasing trend of water spill in September.The volume of irrigation water supply increased significantly for all months.The turbined flow and the hydropower generation increased for all months except May and June, which decreased.Turbined flow increased significantly in January, March and July to December while for hydropower generation, the months that showed significant increasing trends are February, March and August-December.

Correlation between hydro-climatic variables, ENSO and hydropower generation
The results of the Spearman correlation test showed that all correlation coefficients between annual hydropower generation and hydro-climatic variables vary from 0.49 to 0.98 (Table 6).The hydropower generated at the Bagré dam over the period 1993-2012 is very strongly correlated to turbined volume, strongly correlated to irrigation, water level outflow and inflow, Mean basin rainfall (mm             whereas the correlation with mean basin rainfall and lake evaporation is moderate.The results show that with annual increasing inflow and water level, annual hydropower generation also increases.This can be observed in the monthly trends where the simultaneous increase of inflow and water level leads to a significant increase in hydropower generation from August to December (Table 5).
The monthly hydropower generation anomaly is less correlated to the ENSO index (-0.23)(Figure 2) but moderately correlated to water level (0.5) and lake evaporation (0.61).The correlation coefficient between the ENSO index and mean basin rainfall and inflow anomalies is negligible (0.02; À0.05).This indicates the need for further studies on the relationship between ENSO and rainfall variability in West Africa.It shows the necessity to investigate the ENSO indices and classification methods in the prediction of rainfall, inflow and hydropower generation in the study area.
For a detailed investigation of the relationship between monthly hydropower anomaly and the ENSO index, in Figure 3 it can be seen that during El Niño phases in 1997 and 2002, there was a decrease of hydropower generation.In comparison, during La Niña phases in 1999, 2000, 2008-2009and 2011, there was an increase of hydropower generation.During El Niño events of 1995, 2007 and 2010, the hydropower generation was abundant while it fell short during La Niña years of 1996 and 2006.This necessitates a thorough examination of the reservoir management.

Correlation between Ziga dam volume and hydropower generation
The correlation coefficients between annual and monthly hydropower generation and Ziga dam volume are 0.45-0.5,respectively (Table 7).Both annual and monthly hydropower generation at the Bagré dam over the period 2001-2012 are moderately correlated to Ziga volume.The results show that with monthly or annual increasing volume at the Ziga dam, more water could be released and annual hydropower generation would increase at the Bagré dam.increased significantly except for water level, lake evaporation and outflow.This increase, combined with the construction of the Ziga dam and its management change, impacted positively the yearly inflow and hydropower generation at the Bagré dam.However, the monthly trend analysis showed that hydropower generation decreases each May and June due to the significant increase in water withdrawals for irrigation.The results ultimately show that annual hydro-climatic variables can moderately be used to predict hydropower generation.But at the monthly scale, only water level and lake evaporation can be moderately relied on to predict hydropower generation at the Bagré dam.Upstream water management plays an important role in the hydropower generation at the Bagré dam.SONABEL and ONEA (National Water and Sanitation Company) who are in charge of the operation of Bagré and Ziga dams, respectively, should work closely in order to maximize dams' management for hydropower generation, irrigation and water supply.However, as the climate and land-use change simultaneously occur in the basin, further studies, combining impacts of climate and land use/cover changes on the hydropower generated at the Bagré dam, should be undertaken.

Figure 1 |
Figure 1 | Location of the NRB and the Bagré dam in Burkina Faso.

Table 1 |
Spatial and temporal resolution of datasets used in the study

Table 2 |
Comparison statistics of monthly observed rainfall and monthly W5E5 and CHIRPS products at Ouagadougou and Ouahigouya stations

Table 3 |
Modified Mann-Kendall, Pettitt's test and SNHT statistics of annual hydro-climatic variables and hydropower in the NRB