Introduction

Water is an essential natural resource for human well-being, societal development, and ecological health worldwide. Water resources are at the heart of several sustainable development objectives, and their sustainable management is intrinsically linked to the various development sectors (agriculture, energy production, human health, infrastructure, manufacturing, and so on) and to the mitigation of climate change1,2. Climate change is one of the major environmental challenges of the 21st century3and it is the primary driver of adjustments on water resources. With rising global temperatures, the hydrological cycle in many watersheds and regions of the world is accelerating, increasingly threatening the availability and quality of freshwater. The direct and indirect effects of climate change on water resources occur through altering water cycles and increasing the occurrence and intensity of extreme hydrological events such as heavy precipitation, droughts, and floods4,5,6. Thus, climate change combined with other pressures such as rising population, economic growth, and other socio-economic factors, is exacerbating local water availability issues and their spatio-temporal distribution7,8. In developing countries, the issues of water and food security are likely to be fast tracked, especially in regions with precarious climatic conditions, in view of past changes in climate variables such as rising temperatures, changes in precipitation patterns, and increasing of extreme weather events9. According to the Intergovernmental Panel on Climate Change10 fourth assessment report, Africa is one of the most vulnerable continents to the adverse impacts of climate change and variability, a situation which is further aggravated by its low adaptive capacity and the interaction of multiple stresses occurring at various levels.

In West Africa, and particularly the Sahel, annual temperatures have risen more rapidly than the global trend, with an average increase across the whole region approximately 1 °C (1950–2010), the largest warming signals have been in the northern part of Sahel (15–20°N), and mostly in the Western part (Mauritania) and Eastern part (Niger and Chad)11. In addition, the West African Sahel has witnessed significant decreasing trends (from 15 to 35%) in annual precipitation during the Sahelian drought that lasted from the 1970s to the late 1990s12,13,14. This decline in the precipitation has profoundly affected the regimes of large rivers in West Africa (Niger, Volta, Senegal, Gambia Rivers), leading to a deficit of 30 to 60% in the annual average river discharges. This situation translated into declining groundwater levels and accelerated desertification process, resulting to major crises of food insecurity and mass migration, which often leads to conflicts15,16. Since the 1990s, there has been some partial recovery in precipitation in the Sahel with a significant spatial and temporal variability from year to year17,18. Other recent studies19,20,21 found that the contribution of the extreme rainy days in the annual cumulative rainfall has considerably increased.

In the West African Sahel, Burkina Faso is a landlocked country with a surface area of around 274,200 km2 and a population of 20 million22. Burkina Faso, like many other Sahelian countries, is particularly exposed to the repercussions of climate change23. This vulnerability stems from the heavy reliance on water resources to sustain vital sectors such as agriculture and livestock and the water supply to local populations24,25. In Burkina Faso, 80–90% of the working population is involved in agriculture sector, which contributes around 30% of the country’s gross domestic product26,27. However, this agriculture is essentially rainfed, which exposes the agricultural sector to serious threats linked to uncertainties on the starting dates and rainy season duration, and the occurrence of prolonged dry spells28. Several recent studies29,30,31,32 have shown that Burkina Faso is expected to witness a substantial increase in air temperature, and consequently, an increase in potential evapotranspiration (PET). This increase in PET could lead to early drying up of many natural or man-made freshwater bodies (lakes, ponds, dams, and so on), and increasing aridity in at least half of the south-western part of the country33. Thus, to ensure sustainable water management and improve development planning in the region, it is vital to have a thorough knowledge about fluctuations and trends of past and current hydroclimatic variables34,35. The Mouhoun River Catchment (MRC), located in the south-western part of the country, is one of the most important river basins in Burkina Faso with several hydraulic facilities installed (irrigated schemes, developed lowlands under partial control, and water and soil conservation practices). Despite its large hydrographic network, only 23,000 ha of a possible 200,000 ha of irrigation potential have been developed36. The MRC provides water for various purposes such as drinking water, irrigation, livestock watering, hydropower generation, industry and ecosystems. In addition, the MRC is also a transboundary catchment of 96 096 km2 shared by Mali (in northern part) and Ghana (in south-eastern part), thereby providing water and associated ecosystem services to other riparian countries. As in other catchments in the region, the MRC is experiencing an increase in of climate change and variability (dry spells, droughts, episodes of heavy precipitation, floods, and so on), which affect water availability, agricultural production and hydraulic infrastructures25,37.

The objective of the study is to investigate the historical hydroclimatic variability and trends in the Mouhoun River Catchment at the Boromo hydrometric station (MRCB). Statistical analyses of historical hydroclimatic variables were used to evaluate the variability and trends in climate and streamflow at the MRCB. The study outcomes may contribute to sustainable water resources management in a context of climate change and safeguard the food-water security role of the MRCB for local communities.

Results

Analysis of hydroclimatic variability in MRCB

Seasonal and interannual hydroclimate variability in MRCB

Figure 1 shows the seasonal and annual variability of hydroclimate time series, including rainfall (P), potential evapotranspiration (PET) and streamflow (Q) over the period 1961–2020 in the MRCB. From the Fig. 1a, it can be observed that there is an average monomodal pattern, with maximum rainfall occuring in August. Most rainfall occurs during the rainy season from May to October, with greatest concentration between June and September. All the synoptic stations follow this pattern, with Bobo-Dioulasso and Ouahigouya stations showing the highest and lowest rainfall respectively. From the Fig. 1b, the average annual rainfall pattern shows a general downward trend from 1000 mm to 600 mm between 1961 and 1990, followed by a gradual recovery from 600 mm to 800 mm of annual cumulative rainfall after 1990. The Bobo-Dioulasso and Ouahigouya stations showing the highest and lowest annual cumulative rainfall, respectively.

In Fig. 1c, the average monthly PET shows a peak in March (around 200 mm), decreasing to a minimum value of 125 mm in August, corresponding to the wettest month. The Ouahigouya and Dedougou stations, in the northern part of the country have the highest monthly PET values, while the Boromo station in the southern part has the lowest values. From the Fig. 1d, it is observed a stable annual PET pattern on average over the period from 1961 to 2020, with slight trends and fluctuations. The Ouahigouya and Dedougou stations show the highest annual PET values. The Boromo station has the lowest annual PET values, with downward trend observed after 1980.

Figure 1e shows the hydrograph of average monthly streamflow at the Boromo hydrometric station, the outlet of the catchment. A low streamflow period is observed between January and May, corresponding to the dry season. The increase in streamflow starts in June and reaches a peak in September. Thus, the lag time between peak rainfall and peak streamflow in the MRCB is one month. From the Fig. 1f, it can be observed that there is a general downward trend in annual streamflow from 30 mm to less than 10 mm between 1961 and 1990, followed by an increase to 40 mm after 1990. The period 1990–2020 show increased variability in annual streamflow.

Fig. 1
figure 1

Seasonal and annual fluctuations in average rainfall (a, b), potential evapotranspiration (PET) (c, d) and streamflow (e, f) in the MRCB over the period 1961–2020. Panels (a)–(d) show also climatic variables at the synoptic stations (Bobo, Boromo, Dedougou and Ouahigouya).

Full size image

Sensitivity of streamflow to rainfall and potential evapotranspiration

The sensitivities of streamflow (Q) to the rainfall (P) and potential evapotranspiration (PET) in the MRCB are shown in Fig. 2. Figure 2a shows that streamflow is highly sensitive to rainfall (R2 = 0.47), as given by the elasticity coefficient \(\:{\epsilon\:}_{P}^{Q}\:\) of 2.1, indicating that a 1% increase in annual rainfall would lead to a 2.1% increase in annual streamflow. It implies that the volume of the river streamflow increases when rainfall increases. Also, it should be noted that the sensitivity of streamflow to PET is not significant (R2 = 0.002), with an elasticity coefficient \(\:{\epsilon\:}_{PET}^{Q}\) of 0.34, six times lower than that of streamflow to P (Fig. 2b).

Fig. 2
figure 2

Elastic coefficients of rainfall (a), and PET (b) to streamflow in the MRCB over the period 1961–2020.

Full size image

Changes detection in hydroclimatic time series

Statistical analysis of multiple change points has been carried out for the annual time series of P, PET and Q between 1961 and 2020 using the PELT approach. Thus, Fig. 3 highlights the changes identified in the hydroclimatic data over the MRCB. Figure 3a shows the change points in the annual rainfall data on the MRCB, marked by the ratio of change (RC) in the average ranging from − 20% to + 12%. Three (3) significant change points are identified: in 1972 (RC = -19.9%), in 1991 (RC = + 12.2%) and in 2006 (RC = + 10.0%). Overall, between the first two sub-periods, the average annual rainfall decreased from 925.7 mm (1961–1971) to 741.8 mm (1972–1990). The next two sub-periods are marked with a recovery in average rainfall, from 832.0 mm (1991–2005) to 915.3 mm (2006–2020).

In Fig. 3b, the changes in PET are less marked, with the ratio of change ranging from − 6% to + 8%. There are mainly two (2) significant change points: in 1974 (RC = -6.4%) and in 1986 (RC = + 8.2%). Between the first two sub-periods, the average annual PET is decreasing from 2058.5 mm (1961–1973) to 1926.1 mm (1974–1985). Then, the average annual PET increased to 2083.4 mm over the period 1986–2020.

Fluctuations in annual streamflow are very marked, with the ratio of change ranging from − 39% to + 69% (Fig. 3c). Two (2) significant change points were identified: in 1973 (RC = -39.4%) and in 2007 (RC = + 68%). Between the first two sub-periods, the average annual streamflow is decreased from 24.7 mm (1961–1972) to 15.0 mm (1972–1990). After 2007, annual streamflow increased to an average of 25.3 mm, higher than the average value over the sub-period prior to the hydrological drought (1961–1972). The period from 1973 to 2006 with lowest values of streamflow covers the decades of severe droughts (1970s, 1980s) in the West African Sahel. Nevertheless, this period includes years with surplus streamflow, notably in 1994 (32.6 mm) and 1999 (38.7 mm).

Fig. 3
figure 3

Change points in annual time series of rainfall, PET and streamflow over the period from 1961 to 2020 in MRCB. The dashed black vertical lines show the location of change points in annual time series, identified with the Pruned Exact Linear Time (PELT) method38. RC refers to relative changes in mean between the consecutive subperiods. The significance of these relative changes is evaluated using the Mann–Whitney U test at 5% level.

Full size image

Modes of variability in rainfall, potential evapotranspiration and streamflow over the MRCB

Figure 4 shows the modes of continuous variability in hydroclimatic variables (P, PET and Q) over the period 1961–2020 in the MRCB, as depicted by wavelet power spectra shown in Fig. 4d-f. The annual rainfall wavelet spectrum shows significant fluctuations in the 2–4 year band over the periods 1974–1978, 1988–1995, 2001–2005 and 2008–2015 (Fig. 4d). The annual PET wavelet power spectrum shows significant fluctuations in the 2–4 year band from 1975 to 1994, and in the 8–16 year band from 1972 to 1988. The annual streamflow wavelet power spectrum shows significant fluctuations over the periods 1990–2000 and 2008–2014 in the 2–4 year band, and from 1995 to 2010 in the 4–8 year band. It is worth noting that the fluctuations observed in streamflow in the 2–4 year band are consistent with those of rainfall, especially after 1990.

Fig. 4
figure 4

Modes of variability in rainfall, PET, and streamflow over the period 1961–2020 in the MRCB. (a–c) show the standardized values of rainfall, PET, and streamflow, respectively. (d–f) show the continuous wavelet power spectra of rainfall, PET, and streamflow respectively. The thick black contour lines delimit the cone of influence (COI) outside which edge effects distort the signal and, therefore, are not considered in the analysis. Within the COI, significant fluctuations at the 10% significant level against red noise are outlined in thick black contour lines. The colour bar scale at the bottom shows the bias corrected power normalized by the variance.

Full size image

The analysis of wavelet cohence transform (WTC) between the climate variables (P, PET) and hydrological variable (Q) at annual time scale over the period 1961–2020 is shown in Fig. 5. The arrows indicate the phase relationship between P and Q or PET and Q: arrows point to the right (left) when the hydroclimatic time series are in phase (antiphase) or when they are positively (negatively) correlated. Also, arrows pointing up mean that the first variable leads the second by 90°, whereas arrows pointing down indicate that the second variable leads the first by 90°39,40. Figure 5a shows a positive and in phase relationship between P and Q that is very marked in the 2–8 year band. This relationship reveals that rainfall is the primary driver of streamflow at all time scales (from annual to decadal). In Fig. 5b, it appears a weak relationship between PET and Q, limited in the 2–4 year band over the period 1990–2010. Also, a significant quasi-decadal zero-phase from 1972 to 1985 is observed, indicating that the two variables move together over this sub-period.

Fig. 5
figure 5

Wavelet coherence transform between the time series of hydroclimatic variables (P–Q and PET–Q) at the annual time scale over the period 1961–2020. The thick black contours lines denote the 90% confidence level region against red noise, and the thin lines being the cone of influence (COI), outside which edge effects distort the signal and, therefore, are not considered in the analysis. The relative phase relationship is denoted as arrows: negative correlations pointing towards the left or out of phase (180°) relationship; positive correlations pointing towards the right or in-phase (0°) relationship. Arrows pointing upward (downward) indicate that Q is led (Q is leading) by 90° the other variable. The colour bar scale on the right shows the bias corrected power normalized by the variance.

Full size image

Trend analysis of hydroclimate variables

Long-term trend analysis: annual and monthly timescales

The results of the trend analysis of the annual hydroclimate variables over the period 1961–2020 are presented in Table 1. A positive trend (Z > 0) is detected in the rainfall in all synoptic stations and averaged data over the MRCB, except in Bobo-Dioulasso station (Z < 0). The results indicated a statistically non-significant (p > 0.05) decreasing trend in Bobo-Dioulasso station. Upward trends are statistically non-significant for all synoptic stations, except Ouahigouya station.

A positive trend (Z > 0) is found in the annual PET in all synoptic stations and averaged data over the MRCB, except in Boromo station (Z < 0). The results indicated a statistically significant (p < 0.05) decreasing trend in annual PET in Boromo station. Upward trends are statistically significant for all synoptic stations and averaged data, except Ouahigouya station.

A negative trend (Z < 0) is detected in the annual streamflow time series in Boromo hydrometric station (outlet of MRCB), with a downward trend statistically non-significant. This result is consistent with the findings obtained for rainfall. This indicates that the streamflow generated have declined over time in the MRCB.

Table 1 Annual trends of hydroclimatic variables (P, PET, and Q) on the MRCB over the period 1961–2020. The Z-values of the modified Mann–Kendall test and Theil-Sen’s slopes for P, PET and Q are presented.
Full size table

Table 2 shows the results of long-term hydroclimate trend at a monthly time scale over the period 1961–2020. The analysis reveals distinct patterns across the wet season (months from June to October). Analysis of monthly rainfall series indicated a significant upward trend only for the months of June (0.281 mm yr-1) and October (0.169 mm yr-1). Analysis of monthly PET series, on the other hand, exhibited a consistent and highly significant upward trend across all months, with the strongest increases observed in June (0.800 mm yr-1) and July (0.738 mm yr-1), suggesting an increase in atmospheric water demand. It can be observed that, despite PET remains in the wet season lower than in the dry season due to the natural intra-annual variability, its magnitude during wet months has gradually increased over the years. Analysis of monthly streamflow at Boromo hydrometric station only showed a significant increase in July (0.018 mm yr-1) and August (0.024 mm yr-1). Overall, monthly trends observed in streamflow are in strong agreement with those observed in rainfall.

Table 2 Monthly trends of hydroclimatic variables (P, PET, and Q) in the MRCB over the period 1961–2020. The Theil-Sen’s slopes for P, PET, and Q are presented for the months of June to October concentrating the majority of the cumulative annual rainfall.
Full size table

Multi-temporal trend analysis

Contrary to the standard approach that assesses the trend over an entire time series, the multi-temporal approach addresses the time series trend through a combination of subseries. Figure 6 shows the results of multi-temporal trend analysis of annual time series (P, PET and Q) over the period 1961–2020 by considering different starting and ending years.

For the rainfall (Fig. 6a), the analysis reveals two (2) alternating periods of decreasing and increasing trends. Over the period 1961–1990, significant multi-temporal downward trends (slopes < 0) are observed (slopes varying from − 60 to -30 mm yr− 1). Over the period 1980–2020, the multi-temporal trends observed are significant and predominantly positive (slopes > 0), with moderate slopes between 0 and + 30 mm yr− 1.

In Fig. 6b, three (3) significant multi-temporal trends in annual PET are revealed. Over the periods 1961–1985 and 1995–2020, the significant downward trends are detected with slopes varying from − 30 to -10 mm yr− 1 (1961–1985) and from − 20 to 0 mm yr− 1 (1995–2020), respectively. The significant upward trends are noted over the period 1975–1990, with slopes between + 10 and + 30 mm yr− 1.

For the streamflow (Fig. 6c), there are two (2) alternating of decreasing increasing trends. From 1961 to 2005, predominantly downward trends (slopes varying from − 2 to 0 mm yr− 1) are observed, while the significant upward trends with more moderate slopes between 0 and + 1 mm yr− 1 are highlighted over the period 1980–2020. However, there are significant negative trends (slopes varying from − 2 to -1 mm yr− 1) in the period from 2005 to 2020.

Fig. 6
figure 6

Multi-temporal trend analysis for the hydroclimatic variables in the MRCB over the period 1961–2020: (a) P, (b) PET, and (c) Q. Blue (red) cells corresponds to positive (negative) trend slope values, with darker tones indicating higher slope values. The black contour lines delineate periods over which trends are significant (at 5% significance level).

Full size image

Analysis of droughts and wet periods in MRCB

Figure 7 illustrates the temporal fluctuations of dry and wet periods at different time scales (6-, 12- and 24-month) in the MRCB over the period 1961–2020. On a shorter timescale (6-month), the variations of drought indices are more pronounced and marked by an alternation of dry and wet periods, with no clear predominance for the three indices (Fig. 7a-c). The cumulative effects of the SPI, SPEI and SSI become striking and clearer with longer dry (Fig. 7, red colour) and wet (Fig. 7, blue colour) periods when the time scale increases. Thus, at the annual time scale (12-month), for the SPI (Fig. 7d, g), we observe a wet period from 1961 to 1970, followed by a predominantly dry period from 1971 to 2005, then a return to wetter conditions over the period 2006–2020. For the SPEI (Fig. 7e, h), the period 1961–1980 is predominantly wet, followed by a drier period from 1981 to 2010, then a return to humid conditions over the period 2011–2020. For the SSI (Fig. 7f, i), the period 1961–1975 is wet, followed by a predominantly dry period from 1976 to 2008, then a return to humid conditions over the period 2009–2020.

Fig. 7
figure 7

Temporal evolution of the SPI, SPEI and SSI values at different time scales over the period 1961–2020. Red (blue) colours represent dry (wet) periods. Dry periods occur when SPI < 0, SPEI < 0 and SSI < 0. (a, d, g) are the SPI on time scale of 6-, 12-, and 24-month, respectively; (b, e, h) are the SPEI on time scale of 6-, 12-, and 24-month, respectively; (c, f, i) are the SSI on time scale of 6-, 12-, and 24-month, respectively.

Full size image

Shifts in ecohydrological status of the MRCB

The evolution of the ecohydrological status of the MRCB is shown in Fig. 8. It reflects the effects of climate variability on the hydrological response of the catchment. In this study, the period 1961–1970 is taken as the reference period. Thus, there is an initial shift to the left in terms of excess rainfall (\(\:{P}_{ex}\)), which translates as a decrease in streamflow, and towards increased aridity or drier conditions (increase in excess energy \(\:{E}_{ex}\)) over the period 1961–1990. Subsequently, over the period 1990–2010, it can be observed a second shift to the right, marked by an increase in streamflow and a decrease in aridity (increase in excess water \(\:{P}_{ex}\)). Finally, between 2010 and 2020, a third shift is observed, corresponding to a slight decrease in streamflow, as shown by the multi-temporal trend analysis (Fig. 6c), coupled with wetter conditions (decrease in excess energy \(\:{E}_{ex}\)).

Fig. 8
figure 8

Plot of shifts in ecohydrological status in the MRCB over the period 1961–2020. The period 1961–1970, at the beginning of the time series, is taken as the reference. A shift from left to right (right to left) along the Pex axis indicates an increase (decrease) in streamflow. Moving upward (downward) along the Eex axis indicates a shift towards drier (more humid) environmental conditions, that is an increase (decrease) in evaporative demand. The colors of dash lines indicate different shifts.

Full size image

Discussion

In this study, the statistical analysis was performed on a number of hydroclimatic variables in order to understand the trends and variability of climate inputs (P, PET) and hydrological responses (Q) over the MRCB during the last six decades (1961–2020). The results of the change detection analysis show that there is non-stationarity in the annual time series of hydroclimatic variables over six decades (1961–2020) in the MRCB (Fig. 3). Multiple changepoints are identified with significant differences between two consecutive sub-periods (p < 0.05). In the annual rainfall series, three (3) significant changepoints are identified in 1972, 1991 and 2006, and marked by the ratio of change in the average ranging from − 20% to + 12%. The annual rainfall is decreasing over the period 1972–1991 and increasing after 1991. These results are consistent with the findings of some previous studies including41 who reported that, the 1970s and 1980s (especially 1972/1973 and 1983/1984 years) were marked by episodes of intense drought in the West African Sahel with negative impact on the MRCB. Also, a partial recovery of rainfall in the Sahel has been reported, with annual cumulative rainfall still lower than those observed in the 1960s17,42,43. In the annual PET, two (2) significant changepoints are identified in 1974 and 1986, and less marked by the ratio of change ranging from − 6% to + 8%. In the annual streamflow, two (2) significant change points are identified in 1973 and 2007, and marked by the ratio of change ranging from − 39% to + 69%. The range of annual streamflow is much larger than those of annual rainfall and PET. The period from 1973 to 2006 with the lowest values of streamflow covers the decades (1970s, 1980s) of severe droughts in the West African Sahel. Sohoulande Djebou44 found that, on the streamflow series of 8 stream gages selected along the Niger river in West Africa over the period 1961–2012, most of the changepoints were observed between 1981 and 1984. Changes in rainfall and PET may be significantly influencing the timing and volume of streamflow.

The results of sensitivity analysis show that streamflow is highly sensitive to rainfall and less sensitive to PET over the MRCB (Fig. 2). The analysis of wavelet coherence transform shows also a positive and in phase the relationship between rainfall and streamflow that is marked in the 2–8 year band, and weak relationship between PET and streamflow (Fig. 5). The variations observed in streamflow in the 2–4 year band are consistent with those of rainfall, especially after 1990. According to Figs. 5 and 6, some fluctuations in streamflow, especially in the 4–8 year band, are not explained by the changes in rainfall and PET, and may be linked to other factors such as changes in environmental conditions and catchment properties.

The results of trend analysis show positive non-significant upward trends in annual rainfall for all synoptic stations and averaged data over the MRCB, except in Ouahigouya station (+ 3.531 mm yr-1). These results are not consistent with the findings obtained by Dee Longueville et al.45who highlighted a significant downward trend of annual rainfall across all the synoptic stations in Burkina Faso over the period 1950–2013. This difference could be due to the differences in the length of rainfall time series data used in both studies. In fact, the 1950s and 1960s, known as wetter decades in West Africa, could have a major influence on the long-term trend. Obtained results in this study are also not in agreement with the findings of Rouamba et al.46 who reported no trend in annual rainfall over the period 1991–2021 at the Boromo synoptic station. The potential evapotranspiration (PET) in all synoptic stations except in Boromo station and averaged data over the MRCB (+ 1.406 mm yr− 1) is characterized with positive significant upward trends. This upward trend in annual PET could be due to global warming, with a faster increase in mean temperature across the Sahel approximately 1 °C over the period 1950–2010 11. Also, the observed significant increases in PET during the wet months reflect a long-term upward trend in inter-annual PET values, likely in response to rising temperatures and changes in atmospheric conditions. Then, the observed increases in PET could lead to water stress, impacting off-season agriculture and water resource availability.

The results of multi-temporal trends analysis between 1961 and 2020 by considering different starting and ending years, revealed significant and contrasting (downward and/or upward) sub-trends in the time series of rainfall and streamflow with low to moderate changes (Fig. 6). Globally, the distribution of multi-temporal trends over the period 1961–2020 in annual streamflow is consistent with that observed in annual rainfall. Decreases in streamflow observed at the start year of the analysis period (1961–2005) seem to persist over a longer period than that observed in rainfall (1961–1985). This could be explained by the effects of PET trends in the 1970–1990 period. The results of the multi-temporal trend analysis demonstrate that the trend direction and magnitude depend on the length of time series and the position of the sub-series within the whole series.

The study also analyzed droughts and wet periods over the MRCB. The results show that at the longer time scale (24-month), the driest period in the 1980s and the wettest period in the 2010s are observed for all three drought indices (Fig. 7). These findings are consistent to those obtained by Sanou et al.43who showed that in the three climatic zones of Burkina Faso over the period 1961–2020, the 1980s and 2010s were respectively the driest and wettest decades. The positive anomalies of drought indices observed in the 2010s at longer time scales (12- and 24-month) could be due to the intensification of the rainfall regime and an increasing trend in the number of heavy rainfall events in West Africa as reported in previous studies18,19,47.

The evolution of the ecohydrological status of the MRCB over the 1961–2020 period compared to the reference period from 1961 to 1970 revealed three (3) shifts: (i) First, towards a decrease in streamflow coupled with an increase in aridity (between 1961 and 1990); (ii) second, an increase in streamflow higher than the decrease previously observed, together with a general decrease in aridity (between 1990 and 2010); (iii) third, a slight decrease in streamflow coupled with a decrease in aridity (between 2010 and 2020) (Fig. 8). Unfortunately, as mentioned by Yonaba et al.48the picture of shifts in ecohydrological status of the MRCB does not allow to quantify streamflow change, but rather provide qualitative insights regarding whether the reported change is mostly driven by climate.

Findings from this study could contribute to improving our understanding past hydroclimatic variability and trends in the MRCB. This study may be useful to stakeholders in charge of monitoring and management of water resources, and to decision-makers for informing adaptation policies under anticipated climate change given the high sensitivity of the hydrological system (MRCB) to climate variability, especially changes in rainfall.

Materials and methods

Study area

The MRCB lies within latitudes 10°45’ and 14°20′ North and longitudes 1°50′ and 5°30′ West. The catchment outlet is located at the Boromo hydrometric station (11°47′ North and 2°55′ West), occupying an area of 55,660 km2 (nearly one fifth of the national territory). The basin covers four administrative regions (Boucle du Mouhoun, Centre-Ouest, Hauts-Bassins and Nord) of Burkina Faso (Fig. 9). The MRCB belongs to three major eco-climatic zones: the Sudanian zone (annual rainfall higher than 900 mm), the Sudano-Sahelian zone (annual rainfall varies between 600 and 900 mm), and the Sahelian zone (annual rainfall less than 600 mm).

The climate is characterized by an alternating dry and rainy season, with the duration depending to the eco-climatic zone. The dry season lasts about eight months in the north and five to six months in the south; the rainy season lasts from June to September in the north, and from April to October in the south. The average annual rainfall varies between 668.7 mm in the north to 1043.8 mm in the south over the period from 1961 to 2020. The temperatures are increasing from south to north. On average, the annual temperatures range from 15 °C in December to 40 °C in April25.

The relief of the MRCB is quite monotonous with an average slope of 2%. Surface elevation of the MRCB range from 248 m to 695 m above sea level (Fig. 9). The soil cover of the MRCB is heterogeneous and closely linked to geological units which are the basement and the sedimentary, and geomorphological evolution. The MRCB has perennial watercourses and several water storage facilities with a total storage capacity of 1,200 million m349,50. The land cover of the MRCB is essentially made up of wooded savannas, open forests and forest galleries along the permanent watercourses, whose width increases as one goes southward.

The population of the MRCB is estimated at around 3.2 million inhabitants22. The MRCB is a major agricultural region, home to over 50% of the agricultural lands in the country and producing crops such as cotton and rice. Irrigated agriculture downstream of dams is dominated by rice cultivation. Market gardening and maize are cultivated in the dry season, particularly on areas where double rice cultivation cannot be practiced due to insufficient water resources.

Fig. 9
figure 9

Location of the Mouhoun River Catchment at Boromo hydrometric station (MRCB) in Burkina Faso.

Full size image

Data used in this study

For this study, the monthly rainfall (P) and potential evapotranspiration (PET) data from four (4) synoptic stations (Bobo-Dioulasso, Boromo, Dédougou and Ouahigouya) over a 60-year period (1961–2020) were obtained from the national meteorological agency (“Agence Nationale de la Météorologie, ANAM”) in Burkina Faso. The distribution of the synoptic stations is shown in Fig. 9. The Thiessen polygon method is used to calculate the spatial variability of climatic conditions in the MRCB for the variables of interest in this study.

The monthly discharge data for the period 1961–2020 at Boromo hydrometric station in the Mouhoun River Catchment were obtained from the national hydrological services (“Direction Générale des Ressources en Eau, DGRE”) in Burkina Faso. The quality of data is checked before their use in this study. Thus, there are some missing data in the discharge time series. Then, to estimate the missing data, the technique developed by Laborde and Mouhous51 was used through the hydrolab software. The complete discharge data was converted into time series of streamflow, which commonly represents surface water resources.

Analysis of hydroclimatic variability

Sensitivity of streamflow to climate variability

To explore the streamflow sensitivity (Q) to changes in climate variables (P, PET) for annual data in the MRCB, the concept of elasticity was used52. The sensitivity of hydrological systems to the long-term climate fluctuations can be defined as the proportional change in streamflow to the change in some climate variable, such as P or PET53,54. By assuming that climate variables are independent, the streamflow variation was regarded as a function of the change in P and PET. Thus, the coefficients of sensitivity of streamflow to rainfall (\(\:{\epsilon\:}_{P}^{Q}\)) and potential evapotranspiration (\(\:{\epsilon\:}_{PET}^{Q}\)) were estimated from Eqs. (1) and (2) expressed as follows:

$$\:{\epsilon\:}_{P}^{Q}=\left(dQ/dP\right)\times\:\left(\:\stackrel{-}{P}/\stackrel{-}{Q}\:\right)$$
(1)
$$\:{\epsilon\:}_{PET}^{Q}=\left(dQ/dPET\right)\times\:\:\left(\:\stackrel{-}{PET}/\stackrel{-}{Q}\:\right).$$
(2)

where \(\:\stackrel{-}{Q}\), \(\:\stackrel{-}{P}\) and \(\:\stackrel{-}{PET}\)are the long-term average values of Q, P and PET, respectively, and the operator \(\:d\) indicates the difference between the dated and average values.

In this study, the elastic coefficients were therefore determined from the linear regression between the partial derivatives. The streamflow elasticity is usually calculated with annual data. For a given independent variable, the higher the value of its elastic coefficient, the more sensitive streamflow is to the climate variable.

Change detection in hydroclimatic time series

The analysis of hydroclimatic variability also involved detecting any changepoint in the annual hydroclimatic time series. The change point detection approach used in this study is based on the changes in mean, from the Pruned Exact Linear Time (PELT) method proposed by Killick et al.38. This method uses a dynamic programming technique and pruning to further identify multiple change points in a time series. PELT method detects the number of change points and their locations simultaneously. The PELT method is applied by using the cpt.mean function within the changepoint R package55. A constraint of minimum 10 years of segment length (1/6 of each time series length) is applied to avoid the detection of spurious change points38.

After identifying change point in time series, the Mann-Whitney statistical test56 at 5% significance level is used to determine if the means of the two different periods (pre and post the detected change-point) are significantly different from each. To assess the changes between two consecutive sub-periods, pre- and post-change point, the ratio of change (RC), which quantifies the deviation of the mean value of each variable after the change point, was obtained using Eq. (3) expressed as follows:

$$\:RC\left(\%\right)=100\times\:\left({\stackrel{-}{X}}_{post}-{\stackrel{-}{X}}_{pre}\right)/{\stackrel{-}{X}}_{pre}.$$
(3)

Where \(\:{\stackrel{-}{X}}_{post}\) and \(\:{\stackrel{-}{X}}_{pre}\) are the mean values of \(\:X\) variable over the sub-periods post- and pre-change points respectively.

Modes of variability in rainfall, potential evapotranspiration, and streamflow in the MRCB

To understand the variability in P, PET, and Q at various timescales (annual, decadal, and above) over the analysis period (1961–2020), we used the continuous wavelet transform (CWT) for each time series. The CWT is a useful tool for detecting and identifying signals with spectral features or other nonstationary properties. The CWT uses a set of non-orthogonal order 6 Morlet wavelets to generate local wavelet spectra, which provides a good balance between time and frequency localization, giving accurate description of the signal in the spectral-space. The normalized wavelet power spectra, in turn, enable us to discern the prevailing timescales of variability and their temporal progression57,58,59.

Then, to investigate how short- and long-term fluctuations in the climate variables (P and PET) identified by the CWT are propagated to streamflow, we used the wavelet transform coherence (WTC). The WTC analysis allows us to measure the cross-correlation and phase lag between two different time series as a function of time and frequency60. The statistical significance of wavelet power can be assessed relative to the null hypotheses that the signal is generated by a stationary process with a given background power spectrum. The computation of CWT and WTC are carried out at a 10% significance level using the biwavelet R package61.

Hydroclimatic trend analysis

Investigation of long-term trends: annual and monthly timescales

To investigate the long-term changes and the amplitude of change in the hydroclimatic time series (P, PET and Q) for a given time scale (monthly or annual), the non-parametric Mann–Kendall trend test at 5% significance level and the Theil–Sen’s slope estimator were both applied to hydroclimatic time series62,63,64,65. The Mann–Kendall (MK) method is a widely used for detecting a statistically significant trend in climatological and hydrological time series25,44,66,67. The method assumes that the time series values are independent and random. However, since most hydrometeorological time series often have autocorrelation, the outcome of the original test is likely affected. Therefore, in this study, the Modified Mann–Kendall (MMK) trend test was used to remove autocorrelation components in the time series sequences before the assessment of trends and slopes68. This statistical test is applied on annual and monthly data covering the rainy season (JJASO) by using the mmky function within the modifiedmk R package69.

Multi-temporal trend analysis

Many trend analyses of hydrological time series involve trend assessment using all available data from the first to the last year of the record. The multi-temporal approach is an alternative method for trend assessment, which involves trend detection on a range of the sub-series within the available record70,71. In this study, the multi-temporal analysis was used to investigate whether periodic multiple-year alternating dry and wet periods in the river catchment affect trend direction and magnitude. Thus, trends in averaged climate variables (P and PET) over the MRCB and streamflow at the catchment outlet are analyzed recursively for every segment of 10- to 60-year segment within the time series under consideration. However, over each analyzed segment of the multi-temporal trend analysis, the magnitude of change is estimated using the Theil-Sen approach. The significance of trends for the hydroclimatic variables is evaluated using the MMK at 5% level69.

Analysis of drought cycles and wet periods in the MRCB

To investigate dry and wet periods in the MRCB, three (3) standardized drought indices are used, namely the Standardized Precipitation Index (SPI)72the Standardized Precipitation Evapotranspiration Index (SPEI)73 and the Standardized Streamflow Index (SSI)74. These indices are the most widely used to characterize dry and wet periods in terms of rainfall and/or streamflow40and can be calculated at different time scales. In this study, the SPI, SPEI and SSI indices were computed at 6-, 12- and 24-month time scales, based on the averaged climate variables over the catchment P (Mean), P (Mean) – PET (Mean) and Q, respectively. These different time scales are chosen to reflect the impacts of precipitation deficits on water resources and how this translates to different societal impacts. Further details about the steps involved in calculating of these indices are available in40. The SPEI R package75 provides specific functions for computing these indices at time scales 6-, 12-, and 24- month over the period 1961–2020: (i) The spi() function (available at https://rdrr.io/cran/SPEI/src/R/spi.R ) was used to compute the SPI and SSI, based on monthly precipitation and Streamflow data, respectively; (ii) The spei() function (available at https://rdrr.io/cran/SPEI/src/R/spei.R ) was used to compute the SPEI, based on the climatic water balance (precipitation minus PET). In this study, SPI-i, SPEI-i and SSI-i denote the i-month (where i = 6, 12, 24) accumulation period of SPI, SPEI, and SSI, respectively. Positive values of these indices (SPI > 0 or SPEI > 0 or SSI > 0) reflect wet conditions, while negative values (SPI < 0 or SPEI < 0 or SSI < 0) represent dry conditions.

Shifts in ecohydrological status in the MRCB

An approach based on the concept of water-energy balance76 is used in this study to assess shifts in hydroclimatic conditions in the MRCB over six decades (1961–2020) using the period from 1961 to 1970 as the baseline period. This framework states that as the climate change signal affects the hydrological response of the catchment, a tendency in maximizing unused water (defined as \(\:{P}_{ex}\)) and unused energy (defined as \(\:{E}_{ex}\)) occurs, as a depiction of the shift in ecohydrological status of the catchment. The direction of the shift can therefore be explained in terms of: (i) increasing streamflow or water stress (moving right or left along the \(\:{P}_{ex}\) axis, respectively); (ii) increased humidity or drier conditions (moving downward or upward along the \(\:{E}_{ex}\) axis, respectively)48,77.

The excess water \(\:{P}_{ex}\) and excess energy \(\:{E}_{ex}\) are calculated as given by the Eqs. (4) and (5):

$$\:{P}_{ex}=\:\left(P-AET\right)/P$$
(4)
$$\:{E}_{ex}=\:\:\left(PET-AET\right)/PET.$$
(5)

Where P, AET and PET are precipitation, actual evapotranspiration and potential evapotranspiration respectively.

In this study, the time series of AET is based on the water balance of the MRCB. On the basis of hydrological year, the change of the water storage is generally small compared to the other fluxes and can be neglected34. Thus, AET is estimated as the difference between precipitation and streamflow on the annual time scale.