Spatiotemporal variability and change in rainfall in the Oti River Basin, West Africa

Understanding rainwater dispersion in a spatiotemporal context is invaluable toward resourceful water management and a food-secure society. This study, therefore, assessed the variations in rainfall at a spatiotemporal scale in the Oti River Basin of West Africa for observed (1981–2010) and future periods (2021–2050) under the representative concentration pathways (RCPs) 4.5 and 8.5 emission scenarios. Rainfall data from meteorological stations and Climate Hazards Group Infrared Precipitation with Stations (CHIRPS) were used. The percentage changes in rainfall for the peak month as well as for rainy and dry seasons under the two climate scenarios were determined. The coefficient of variation (CV) and the standardized anomaly index (SAI) were used to assess annual variations in rainfall. In general, under both emission scenarios, rainfall is projected to decrease over the study area. However, the amount of rainfall during the peak month (August) for RCP4.5 and RCP8.5 could increase by 0.26 and 9.3%, respectively. The highest SAIs for the observed period were þ1.58 (2009) and 2.29 (1983) with the latter showing a relationship with historic drought in the basin. The projected SAI under RCP4.5 and RCP8.5 indicated extremely wet (þ2.12) and very wet (þ1.91) periods for the years 2037 and 2028, respectively. The study provides relevant information and a chance to aid the design of innovative adaptation measures toward efficient water management and agricultural planning for the basin.


INTRODUCTION
Sustainable management of water resources, agricultural production and all hydro-ecological services at a river basin level is directly linked to empirical knowledge of the distribution of rainfall in space and time over the basin. Understanding climatic conditions, particularly spatiotemporal distribution of rainfall, is therefore vital for natural resources management in developing countries as in Africa with rising population growths and weak responsive capacity to climate change and variability impacts (Ampadu 2021).
With significant rainfall variability in time and space, West Africa is sensitive to extreme droughts and flooding due to climate change (Lebel et al. 2009). This is due to an undeviating relationship between rainfall and global climate (Twisa & Buchroithner 2019;Nhemachena et al. 2020). The variability of rainfall has caused lengthy droughts and floods in most parts of West Africa, posing a serious threat to the region. The savannah and semiarid areas, since the 1960s, have recorded events of famines and floods (Epule et al. 2017). Regrettably, climate variability is likely to intensify (Suleiman & Ifabiyi from CHIRPS, as presented in Table 1 and shown in Figure 1. Daily rainfall CHIRPS data were extracted in R software (Packages: ncdf4 and raster) applying respective locations of the 22 gridded points for 1981-2010. Several research studies (e.g., Dembélé et al. 2020;Muthoni 2020;Satgé et al. 2020) have applied CHIRPS in both the Volta basin and the West African region and have proven that it can accurately estimate station data, and therefore the CHIRPS data can be used in place of station data.

Climate models' datasets
The study used rainfall simulations from global circulation models (GCMs) under the Coordinated Regional Climate Downscaling Experiment (CORDEX-Africa) (Samuelsson et al. 2011;Kjellström et al. 2016). The GCMs were downscaled by the Rossby Centre regional atmospheric model (RCA4) at a spatial resolution of 0.44°Â0.44°(∼50 kmÂ50 km). Table 2 shows the climate models and the institute that created them. The GCMs-RCMs were selected because they have been widely used across the Volta basin (e.g., Kunstmann & Jung 2005;Annor et al. 2017). In addition, Akinsanola et al. (2017) and Agyekum et al. (2018) used these GCMs to evaluate rainfall simulations over West Africa and have attested to their suitability. The study considered climate change datasets under representative concentration pathways (RCPs) 4.5 and 8.5 for the future period 2021-2050.
To evaluate the efficiency of the RCMs, the simulated historical rainfall was compared with the observed rainfall at the mean monthly scale using the Taylor diagram. The performance evaluation was done at a monthly scale because it best shows the characteristics of rainfall change (Gulacha & Mulungu 2016). The commonly used statistical measures like Pearson's correlation coefficient (r), standard deviation and root-mean-square error (RMSE) were employed to evaluate the Journal of Water and Climate Change Vol 00 No 0, 3 Uncorrected Proof effectiveness of the RCMs. The position of each model and the ensemble mean in the acceptable range of the statistical measures [(r¼À1 to 1) and (RMSE¼0-1, with 0 being the perfect fit)] showed how well it simulates rainfall in the basin (Moriasi et al. 2007;Dembélé & Zwart 2016;Bessah et al. 2020).
The simulated daily rainfall dataset in this study was extracted and bias-corrected using the CMhyd application. The bias-correction method employed was the quantile-quantile mapping technique (Boé et al. 2008;Johnson & Sharma 2011). The CMhyd took the observed and the simulated rainfall data (thus, the historical and future scenarios) and used the overlapping period 1981-2005 to calculate the correction parameters for the future period (2021-2050) under RCP4.5 and RCP8.5 scenarios. The bias-correction technique helps to adjust the simulated RCM climate values to fall in line with the observed (Teutschbein & Seibert 2012). The procedure for the bias-correction as demonstrated in Thom (1958) is as follows:   for the day d, with month m, P Ã scen is the simulated rainfall under historical climate scenario, F g is the Gamma cumulative distribution function, α is the shape parameter and b obs,m is the scale parameter for the month.

Rainfall trend and variability analysis
To explore the occurrence of rainfall in the ORB, the monthly rainfall analysis was conducted for both the observed  and future periods (2021-2050) under RCP4.5 and RCP8.5 scenarios. The percentage change in rainfall for the peak month (August) under the future scenarios was computed relative to the observed. Also, the amount and rate change in rainfall for rainy and dry seasons were determined.
Variations and trends in rainfall at an annual scale for the historical period  and the future period (2021-2050) under RCP4.5 and RCP8.5 scenarios were analyzed using the coefficient of variation (CV) and the standardized anomaly index (SAI).
The CV was computed using the following formula: where σ is the standard deviation and μ is the mean rainfall for the temporal scales used. The degree of rainfall variations were low (CV,20), moderate (20,CV,30) and high (CV.30) (Alemu & Bawoke 2019). The SAI of rainfall was calculated using the following equation: where X i stands for the yearly rainfall for the period understudy; X represents the long-term average yearly rainfall for the observation period and σ corresponds to the standard deviation of yearly rainfall during the observed period (Alemu & Bawoke 2019). Table 3 shows the SAI value classification used.

Mann-Kendall trend test, Sen's slope estimator and projected changes
The study employed the Mann-Kendall (MK) trend statistics to assess trends in rainfall, and Sen's slope estimator was also used to determine the magnitude of the trends over the ORB. The MAKESENS Software was used to compute the trends at a 5% significance level. The test is universally acknowledged due to its robustness, least susceptible to outliers and appropriate for discovering tendencies in time series records (Siraj et al. 2013). It is the most commonly recommended test, and it has

RCMs' performance and bias-correction
The result obtained for comparing the RCMs with the observation is demonstrated on the Taylor diagram in Figure 2. It can be seen that the correlation values obtained from the comparison are above 0.9, with a standard deviation value below 2.0 and an RMSE value below 0.75. The CNRM-CM5 model had the strongest correlation (R¼0.99), the least centered RMSE (below 0.25) and a lowest standard deviation (below 1.25). Although the GFDL-ESM2M model was found to be the weakest, it recorded a correlation of 0.93, a centered RMSE below 0.75 and a standard deviation below 1.75. Although the CNRM-CM5 model performed best among the individual models, the models' ensemble also performed better than most single models with a correlation of 0.98, RMSE below 0.25 and standard deviation below 1.25. The findings are similar to those of Akinsanola et al. (2017) and Lin et al. (2020) who also noted that models' ensemble mean makes the representation of rainfall characteristics better than the majority of single models. Generally, the results show a close match between the RCMs and ensemble simulated rainfall pattern and observation. Figure 3 provides the result from the bias-correction of the eight CORDEX-Africa RCMs. The correlation between the raw models' ensemble and observation were 0.76 and the NSE value of 0.53. The uncorrected models each show clear variances in how it reproduces historical rainfall (Figure 3(a)). However, after bias-correction, the individual models and their ensemble are seen to be enhanced (Figure 3(b)). After bias-correction, a correlation between the ensemble and observed was 0.91 with an improved NSE (0.98). Generally, it is observed that the corrected individual RCMs and their ensemble were improved to reproduce the rainfall pattern demonstrated by the observed records at the monthly scale. Hence, the bias-corrected RCMs can be said to represent the ORB and are considered reliable for the analysis.
Month-to-month distribution of rainfall Figure 4 illustrates the annual cycle of rainfall per month for both observed  and near-future periods (2021-2050) under RCP4.5 and RCP8.5. It is observed that the monthly rainfall for the future period would have its peak in August just like the observed period. The amount of rainfall for the peak season, August, is revealed to be 246.1 mm for the observed period; 246.8 mm (about 0.26% peak increment) and 269 mm (about 9.3% increment) of rainfall in the near-future period for RCP4.5 and RCP8.5, respectively. The expected rise in rainfall during the peak month could cause the basin to experience some flooding episodes, and great disturbance to agricultural production (crops and livestock). It can also lead to the outbreak of rainrelated diseases like malaria. It is worthy to note that the basin's adaptation and mitigation efforts be strengthened during this period so as to help curb the occurrence of any devastating situations. Uncorrected Proof The amount of rainfall for the rainy season (April-October) (Klassou & Komi 2021) is expected to be around 947.7 mm (a decrease of about 8.8%) under the RCP4.5 scenario and around 1,004 mm (3.4% decrease in rainfall amount) for the RCP8.5 scenario comparing with the observed that recorded 1,038.8 mm of rainfall. For the dry season (November-March), the  Uncorrected Proof amount of rainfall is expected to be around 22.5 mm (a decrease of about 36.4%) under RCP4.5 and about 24 mm (a decrease of about 32.3%) for the RCP8.5 scenario in relation to the observed rainfall of 35.4 mm. As water is crucial to crop production in the basin and agriculture in the region being highly rainfall-dependent, the projections of future rainfall change could greatly influence crop production (Reilly et al. 2003;Olesen et al. 2007;Gornall et al. 2010).

Annual rainfall distribution
Mean, standard deviation (SD) and CV were computed for 1981-2010 and 2021-2050 (Table 4). During the observed period, mean annual rainfall was 1,073.9 mm which ranged between 783.4 mm at GRID3 and 1,464.8 mm at GRID20. The mean annual rainfall during the 2021-2050 period recorded 970.2 mm under RCP4.5 which ranged between 640.2 mm at GRID3 and 1,408.3 mm at Sokode; and 1,027.9 mm under the RCP8.5 scenario which also ranged from 758.4 mm at GRID3 to 1,287.1 mm at Sokode. It is observed that the lower basin receives a high amount of rainfall as compared to the middle and upper basins during the observed and future periods. Although rainfall in the lower basin would rise in the future period, the amount would be reduced. The study discovered a reduction in the mean annual rainfall for the future period under both emission scenarios. This could reflect the predictions of the IPCC (2007) that subtropical areas would experience a decreased rainfall due to the changing climate. Figure 5 illustrates the allocation of rainfall spatially at the basin for observed and future periods. It is generally seen that rainfall in the basin is more pronounced in the lower basin as compared to the upper basin. Rainfall would generally decrease entirely across the basin under the RCP4.5 scenario. However, under RCP8.5, rainfall in the lower basin would further decrease, while the upper basin would experience a slight recovery of rainfall. Similarly, rainfall is expected to decline in future under both climate scenarios at a temporal scale (Figure 6). From the time series, it is detected that annual rainfall in the basin during the historical period was between 835.5 mm (1983) and 1,239.1 mm (2009). In the future period and under RCP4.5, annual rainfall is expected to range between 842.4 mm (2027) and 1,082.3 mm (2037); and between 900.6 mm (2041) and 1,126.1 mm (2028) of rainfall for RCP8.5. Declined rainfall anticipated at both spatial and temporal scales could severely impact agricultural planning and crop production in the basin. In addition, it could greatly impact streamflow and water levels in the basin. Subsequently, the existing adaptation strategies in the basin must be evaluated in order to ensure that the livelihoods of peripheral communities are not hindered by expected changes.  Table 5 show the projected changes in rainfall at the ORB. For RCP4.5, almost all stations would experience a reduced amount of rainfall except for the Sokode station which recorded an increment. Similarly, under RCP8.5 in the future, the majority of the stations in the basin would experience a reduced amount of rainfall (particularly stations in the lower Uncorrected Proof basin), except for a few stations which would record an increment in rainfall. Rainfed agriculture in the West African subregion could significantly be at risk (Owusu & Waylen 2009). In addition, Lare & Nicholson (1994) have pointed out that inadequate large-scale rainfall, as shown at the stations, can affect ecosystem. The MK test for annual rainfall in the basin   . The MK test for the basin during the future period (2021-2050) under RCP4.5 exposed an insignificant increasing trend at a 5% significant level. However, the Niamtougou station revealed a significant increasing trend. Rainfall across the basin revealed an insignificant decreasing trend in the future (2021-2050) for the RCP8.5 scenario ( Table 6).

Projections in annual rainfall
Variability of rainfall for observed  and future periods  The standard deviation ranged between 94.8 mm at GRID1 and 320.9 mm of rainfall at Kete-Krachi with a mean value of 104.1 mm. CV varied from 10.2% at GRID16 to 23.6% at Kete-Krachi with a mean CV of 9.7%, indicating less rainfall variability (,20%) according to Asfaw et al. (2018) and Alemu & Bawoke (2019). The moderate variability in rainfall at Kete-Krachi (23.6%) and Dapaong (20.3%) suggests that the amount of water available in these areas was somewhat more erratic compared to the areas with low CV (Alemu & Bawoke 2019). The mean standard deviation recorded under RCP4.5 was 52.8 mm for the basin, which varied from 56.4 mm at Fada to 100.9 mm at Niamtougou. Under the RCP8.5, the mean standard deviation is 51.4 mm for the basin, ranging between 54.5 mm at GRID3 and 91.2 mm at Sokode. The mean CV in the basin recorded for RCP4.5 was 5.4% ranging from 5.4% at GRID19 to 9.7% at GRID3. The mean CV in the basin under the RCP8.5 scenario recorded 5% which ranged between 5% at GRID16 and 9% at Natitingou. This shows a low variation of rainfall for RCP4.5 and RCP8.5 scenarios. Although almost all stations revealed a low to moderate variation in rainfall amount for the understudied periods, studies such as Nyatuame & Agodzo (2017), Ayanlade et al. (2018) and Alemu & Bawoke (2019) have stated that such variations can adversely impact agriculture and water management. The spatial distribution of CV for annual rainfall is given in Figure 8. The CV for the entire basin was found to be high during the historical period but shows a general increase from the lower basin to the upper basin.
Anomalies of rainfall for observed  and future periods  The annual rainfall anomalies in the ORB for the observed  and future periods (2021-2050) under RCPs 4.5 and 8.5 scenarios are shown in Figures 9 and 10, respectively. The results exposed the year-to-year variations in rainfall across the basin. The year 2009 is seen to have had the highest positive anomaly (þ1.58), while 1983 had the highest negative anomaly (À2.29). This conforms with the study by Kasei et al. (2010) which found out that 1983 was the driest year in the Volta Basin.
More than 90% of the basin was in a severe state of drought and a moderate drought occurred prior to 1982. From the results, the early 1980s experienced pronounced negative anomalies. This shows a relationship with the historic drought in the basin and West Africa between 1961 and 2005 (Kasei et al. 2010). This resulted in the loss of lives and farm animals in desertlike areas of Burkina Faso (Dembélé & Zwart 2016). Generally, the historical period had 4 very wet years (1991, 1994, 2003 and 2009), 1 moderately wet year (1999), 5 moderately dry years (1982, 1984, 1990, 2001 and 2006), 1 extremely dry year (1983) and the remaining 19 years being near normal years. The SAI for the future period is displayed in Figure 10. In the future period, the negative anomaly would be more pronounced in 2027 (À2.42) under RCP4.5 and in 2041 (À2.48) under the RCP8.5 scenario, indicating a possibility of extremely dry condition. The positive anomaly is expected to be more pronounced in 2037 (2.12) under RCP4.5 and 2028 (1.91) under RCP8.5, indicating extremely wet and very wet periods, respectively   Note: values in bold are significant at 5%; S. slope¼Sen's slope.

Implications of projected variation in rainfall on land use and water yield
Cropland is seen to be the dominant land use in the ORB from the 2016 land-use land-cover map ( Figure 11). Farming covers areas from latitude 9°N to the northern part of the basin which was observed to have an annual rainfall between 800 and 1,300 mm ( Figure 5). The projected changes under RCP4.5 (decrease between 20 and 180 mm) could make the forest areas below latitude 9°N also suitable climatologically for farming if soil and other factors are conducive (Travis 2016). Moreover, the projected decrease in the upper basin may influence a shift or gradual migration of cropping areas toward the lower basin. On the other hand, the projected increase in annual rainfall under RCP8.5 from the middle basin toward the upper  Uncorrected Proof basin could result in changes of cropping systems that make use of the additional rainfall amount (Chemura et al. 2020). Also, it might imply that more water resilient varieties would be needed. Grassland and cropland may be expanded to meet up with demands of food and livestock feed as climate change adaptation or coping strategies in the ORB. This means that as an exchange for higher agricultural productivity in the basin, the conversions to cropland can affect the ecosystem services predominantly, especially water yield due to increased runoff (Brink & Eva 2009;Li et al. 2018). Changes in land use, particularly the continuous expansion of cropland, have contributed to the degradation of vegetation and water scarcity in recent years, as reported elsewhere (Sajikumar & Remya 2015;Woldesenbet et al. 2017;Balist et al. 2022). Land-use/ land-cover changes alter the rainfall path into runoff by affecting important hydrological elements like surface runoff, groundwater recharge, infiltration, interception and evaporation (Song & Deng 2017;Woldesenbet et al. 2017;Balist et al. 2022).
Water yield is one of the most important ecosystem services, and it is critical to the regional economy and ecosystem's long-term sustainability (Yang et al. 2021). The spatial distribution of CV of rainfall in the basin (Figure 8) shows that extreme changes may not occur and, therefore, could prevent sudden migration of farm activities from one location to the other. As the ORB is also crucial to the West African sub-region's economy, it is obvious that the expansion of cropland and decline in tree cover may become vital environmental stressors and subsequently impact surface runoff and water yield in the basin. In addition, the anticipated decline in rainfall in the basin during the near-future period (2021-2050) under both RCP4.5 and RCP8.5 scenarios calls for an evaluation of land use and climate impacts on water yield in the basin in order to know the future water availability.

CONCLUSIONS
This work has investigated rainfall variations and anticipated change over the ORB for observed  and future periods (2021-2050) under RCP4.5 and RCP8.5, using data from meteorological stations, CHIRPS gridded-observed data as well as CORDEX-Africa. Percentage change in rainfall for the peak month as well as for rainy and dry seasons under the two climate scenarios was determined at a monthly scale. The mean annual rainfall in the ORB was 1,073.9 mm, SD of 104.1 mm and CV of 9.7% for the observed period. The annual average rain in the future period under RCP4.5 recorded 970.2 mm, SD of 52.8 mm and CV of 5.4%. Under RCP8.5, the basin recorded 1,027.9 mm average rainfall which had SD and CV of 51.4 mm and 5%, respectively. Generally, rainfall amount would decline under both emission scenarios. At a monthly scale, percentage rainfall increment for the peak month (August) is expected to be around 0.26% for RCP4.5 and 9.3% under RCP8.5. The amount of rainfall for the rainy (April-October) and dry seasons (November-March) would experience a decrease under both climate scenarios. Also, during the observed period, the highest positive anomaly of 1.58 was seen in 2009, with the highest negative anomaly (À2.29) recorded in 1983 revealing a connection with historic drought conditions. The year 2037 would experience a more positive anomaly (2.12), while a high negative anomaly (À2.42) is expected to be recorded in 2027 under RCP4.5 disclosing a possible very wet and extremely dry situation. Also, a high positive anomaly in the year 2028 (1.91) and a high negative anomaly (À2.48) in the year 2041 under RCP8.5 are observed unveiling a very wet and extremely dry condition to be expected, respectively. The trend analysis presented both statistically nonsignificant growing and declining movement for RCP4.5 and RCP8.5, respectively, with the historical period revealing a significant Uncorrected Proof increasing trend. The expected decrease in rainfall is cause for concern because it signifies that the dependent countries' source of livelihood and economic boost could be endangered. In addition, the expected moderately dry, severely dry and extremely dry years in the future could throw off the balance between water demand and availability, putting basin-dependent countries in a water stress situation. Although the study relied on gridded datasets to complement missing data, its finding offers crucial information for a better understanding of the spatiotemporal distribution and variations necessary for the sustenance of economic development and livelihoods through such sectors as agriculture, water resources, ecosystem and biodiversity. However, the impact of projections on land use and water yield could only be inferred. Therefore, modeling the basin hydrology for the accessibility of water in the future is advised. Research would be necessary to assess the impacts of land-use and land-cover changes on rainfall variability and change over the basin. Additionally, the future study should concentrate on climate change resilient management strategies in the ORB.