Evaluation of the Climate Variables
The ensemble mean of models (referred as RCMEAN) appropriately represents, over West Africa, moisture variables such as precipitation (PRE), the potential evapotranspiration (PET), and the climate water balance (CWB = PRE-PET) with a very good and significant (99% of confidence level) correlation (\(r \ge 0.90\)) compared to observed CRU. The precipitation is well reproduced regarding the observed CRU. The model was able to capture the spatial gradient of precipitation over the study area, with well-located maxima (maximum around the Gulf of Guinea and minimum in the Sahel), with some scattering location of the maximum rainfall in southern of Nigeria, Guinea-Conakry, and Liberia, and the west of Cameroun. In terms of amount, the bias pattern shows that RCMEAN globally fairly overestimates the precipitation except the southeastern part of the study domain where the model displays an underestimation of the precipitation. From the assessment of the climate water balance (Fig. 2g–i), it can be seen a very important water deficit (negative bias) in the Savanna and Gulf of Guinea, which is due to the underestimation of precipitation, while in the Sahel, there is a surplus of water. However, particularly high values are recorded around coastal countries as Liberia, Sierra Leone, and southern Nigeria.
The inconsistency of the climate water balance assessment between the simulated models and observed can be assigned to various factors. For instance, the wet (positive values) bias over western part of the study area indicates that the convective parameterization schemes at 0.44º horizontal resolution in RCA4 model may be too active in producing precipitation over this area (Abiodun et al. 2018), while the dry (negative values) bias over the eastern part of the study area suggests that the convective parameterization schemes are not fully resolved over the eastern area, producing less moisture available for inland rainfall. It can also be due to the density of the weather stations available over the area (both Eastern and Western) for the observed data estimation. The method of calculating PET can also be a factor of the disparity. Abiodun et al. (2018) assessed the uncertainty of PET estimation with Hargreaves and Penman methods and found that its uncertainty contributes to the CWB’s divergence.
The annual cycle of precipitation presented in Fig. 3 lies within the model ensemble spread in all the sub-zones, and the ensemble mean closely follows the observed curve. In the Gulf of Guinea, RCMEAN and CRU indicate two rainy seasons, while the Savannah and Sahel have shown a single-mode diet with a dry season (dry winter) and a single rainy season (wet summer). These observations reflect the seasonal fluctuations (oscillation) of the ITD over West Africa. The average value of precipitation over the Gulf of Guinea and Savanna is increasing from the second part of May up to a peak (180 mm month−1 for the Gulf of Guinea and 230 mm month−1 for the Savanna) in August when the ITCZ reached its most northerly position (second quasi-stable position) about 10ºN. The average rainfall value in the Sahel also peaked in August, but showed a later increase (July) compared to other regions. The driest period is from October to March in the Gulf of Guinea and from October to May for the Savanna, where the PET increases till reaching its maximum value. It can also be noticed that during the rainy season the PET value dropped to its minimum value. The observed value does not lie within the model spread but follows the model’s curve and underestimates the simulations over the Gulf of Guinea and Savanna. The Sahel is dry for the whole year because the CWB is negative for the historical period. The PET here agreed well by lying within the models and follows the curve of simulations.
According to the good agreement of the models with respect to the observed (Figs. 2, 3), the evaluation of the spatial pattern to detect potential drought and flooding areas is performed using the SPEI for various scales to focus on different types of droughts. The SPEI1 is used to characterize the meteorological drought, while the couple (SPEI3, SPEI6) and (SPEI9, SPEI12) are referred to agricultural and hydrological droughts, respectively. Figure 4 shows that RCMEAN reproduces the patterns in phase opposition (with negative correlation). This opposite performance may be due to various factors. It can be related to the temporal gridded average for each model and the ensemble mean of models. However, it has to be kept in mind that on a grid, the SPEI has either negative or positive values and its averaging could be the reason for the misrepresentation of both model ensemble mean and observed. It can also be mentioned based on Fig. 4 that the more the SPEI scale increases, the more the model improves its reproducibility with the observed pattern. Another factor can be the potential large discrepancy among the simulated patterns.
Assessment of the Extreme and Severe Dry Events for Historical Period
To explain the misrepresentation of ensemble mean SPEI pattern with respect to CRU (Fig. 4), the computation of the magnitude of drought frequency is undertaken. Figures 5 and 6 show, respectively, the frequency of extreme (SPEI < − 2) and severe (− 2 < SPEI < − 1.5) droughts in West Africa for the historical period for both the observed and the RCMEAN. The performance of the model in simulating drought intensity and frequency over the study area depends on the scale of which the SPEI is computed. The model reproduces the extreme and severe droughts for each type of drought with positive correlation. The agreement of the model decreases when the scale of the SPEI increases (pointing out with the correlation r computed between the RCMEAN and CRU), which means that the RCMEAN captures the meteorological and agricultural extreme and severe droughts better than the hydrological extreme drought. However, the ensemble mean at the scale of meteorological and agricultural droughts overestimate the frequency of extreme drought events up to 2 events per decade at the northwestern part and the Gulf of Guinea. The underestimation of the frequency of extreme drought lies in the Gulf of Guinea for both agricultural and hydrological drought events. Nevertheless, the models do capture well the magnitude of hydrological extreme drought over Nigeria, Benin, south of Ghana, and northern part of Niger with a bias close to zero. According to the model, the high values of the frequency in extreme drought are located at the Gulf of Guinea, the Sahel, and the eastern part of the study area including Chad and the north of Nigeria for the agriculture drought. Conversely, the model records the low value of the frequency of agricultural extreme drought in Niger, Ghana, Cote d’Ivoire Guinea, and Cameroon. It shows that the frequency of severe drought is from 4 to 8 events per decade, while for the observed this frequency is between 2 and 12 events per decade. Globally at the scale of meteorological drought, there is an overestimation of the frequency of severe drought except for countries as Mauritania, Mali, and Cameroon, which present an underestimation of 2 events per decade in response to the frequency to CRU. The model at the agricultural scale underestimates the severe drought over Chad, northern Nigeria, southern or Benin, Niger, and Mauritania up to 4 events per decade. The model fails to reproduce the hydrological severe drought; it widely underestimates the frequency of severe drought in Niger, Mali, Mauritania Nigeria, Chad, Benin, southern Ghana, and Cote d’Ivoire. The highest value for its misrepresentation is over northeastern Nigeria (a part of Lake Chad) and Mauritania, and northern of Chad where the model evaluates the magnitude of the hydrological severe drought about 7 events per decade against 12 events per decade according to the observation. The model for all types of severe drought overestimates the magnitude over Senegal, Mauritania, eastern of Mali, and the northern part of Niger, Cote d’Ivoire, and Ghana up to 4 events per decade.
Assessment of the Projection of Extreme and Severe Dry Events
Figures 7 and 8 present the variability of extreme and severe dry events, respectively, under different GWLs and various drought types. Figure 7 indicates that more than 80% of RCA4 models under the GWL1.5 °C show that West Africa experiences a northward increasing trend of extreme dry events (i.e., there is an increase in the frequency of extreme drought events compared to the historical period from the south to the northern part) which is materialized with the black cross (+). Conversely, for the GWL1.5 °C a southward decreasing trend of severe drought is recorded (proved with the horizontal black stripe in Fig. 8), and the frequency of the event increases from the meteorological drought to hydrological drought. In terms of frequency, over the Sahel, the extreme dry weather frequency is above of 4 events per decade and about 3 events per decade in the Savanna. In the Gulf of Guinea, this frequency is about 2 events per decade. In the meteorological drought for the other GWLs (2.0 °C, 2.5, and 3.0 °C), the Gulf of Guinea and Savannah experienced an increase in extreme drought trends with a frequency of about 2 events per decade, while over Sahel and Savanna more than 80% of the models show a downtrend of the extreme dry event for agricultural and hydrological droughts. This decrease extends also to the Gulf of Guinea for the SPEI9 under GWL2.5. Only some coastal countries such as Ghana, Cote d’Ivoire, and Cameroon experience an increase in the extreme dry events during the projected periods. Globally, the frequency of the extreme drought is between 1 and 2 events per decade for the GWLs 2.0 °C, 2.5 °C, and 3.0 °C. However, it is projected a uniform pattern of extreme drought events over West Africa according to the RCA4 models.
In summary, under the interested GWLs used in this study, much variability has been noticed with the reference period. The ensemble mean of models shows either important extreme or severe dry events. Globally, under the GWL1.5 for all drought types studied, a recurrent increase in extreme dry events is noticed, particularly the Gulf of Guinea and Savanna for both events experienced an increasing trend, while the Sahel illustrates an increase in only extreme dry events. For the GWL 2.0, 2.5, and 3.0, at agricultural and hydrological drought scales, a high important decrease in the extremely dry events is perceived over the Savanna and the Sahel particularly around countries as Niger, Mali, Burkina Faso, Benin, and Nigeria. The coastal countries detect an increase in extremely dry events, while the southeastern area notes an increase in the extremely dry events. To figure out the causes of these various drought events recorded, an analysis on the concentration of the precipitation according to the simulations of CORDEX-RCA4 is undertaken.
Annual Variability of the Precipitation Concentration Index
The annual scale of the PCI calculated in this study varies generally across the study area from values greater or equal to 10, to higher than 20; according to Oliver’s (1980) classification, this denotes a seasonal rainfall regime. Figure 9 presents the variability of PCI on annual, rainy season (wet season), and dry season periods over West Africa for the GWLs 2.0 °C and 3.0 °C. The display of PCI of GWLs 1.5 °C and 2.5 °C (not shown) is similar to the analysis of GWLs 2.0 °C and 3.0 °C. From Fig. 9a–c, the lower values recorded during the historical period (here called the control period or CTL) are between 10 and 13 on the Gulf of Guinea, thus illustrating a moderate precipitation concentration over this area. The seasonality is more pronounced in the transition area (the Savanna) with a PCI between 17 and 18, which shows how the precipitation concentration is irregularly distributed; lastly, the Sahel area has a high precipitation concentration (PCI>20), which means that the precipitation is strongly and irregularly distributed.
For the different GWLs studied, it is noticed that, for the Gulf of Guinea and the Savanna, an irregular precipitation concentration exists, except for some countries (Liberia and Côte d’Ivoire), which have a low precipitation concentration, while a strong irregular precipitation distribution is observed in the Sahel.
Seasonal Variability of the Precipitation Concentration Index
The PCI calculated for the seasonal scale shows complex spatial patterns of precipitation distribution in the area of study. Thus, Fig. 9d–f illustrates the uniform precipitation concentration (i.e., PCI values below 10, Oliver 1980) across the Gulf of Guinea and the Savanna. For the specified GWLs, the average of the uniform precipitation distribution extends toward the Sahel. The northern part of the study area records an irregular precipitation concentration (16 ≤ PCI ≤ 20) during the wet season.
Figure 9 (g–i: PCI for the dry season) shows that, during this period of the year selected, an irregular precipitation concentration is only observed over the Gulf of Guinea. All the other areas, such as the Savanna and the Sahel, have a strong irregular precipitation concentration, which means that the total precipitation occurs within 1 or 2 months.
The results from Fig. 9 (concerning the annual and seasonal evaluation) confirm that the precipitation in West Africa is uniformly distributed during rainy season in the Gulf of Guinea and the Savanna. Despite the global warming effect for all levels, this precipitation concentration does not change; on the contrary, it extends toward the Sahel. In general, the highest values of PCI are recorded over the Sahel, whereas the lowest occur over the Gulf of Guinea.
Variability of the Precipitation Concentration Degree and the Precipitation Concentration Period
Figure 10 displays the PCP and PCD. Figure 10a–c illustrates that the range of PCPs across West African region for the present or future study periods is around 7 ± 2. This means that the yearly mean PCP on West Africa is from June to September, a period of rainfall production governed by the West African Monsoon (WAM). The highest value for the historical period (1971–2000, also referred to as the control period) is recorded over the northwestern part of West Africa, while for the projections, this value is located over the Sahel. The result confirms that the rainy season arrives earlier in the southern areas, followed by the transition area (the Savanna), before reaching the Sahel. The mean yearly PCDs (Fig. 10d–f) vary from 0.17 to 0.90, denoting the high variability of the precipitation concentration over West Africa. During the present period (Fig. 10d), the PCD values increase, suggesting the existence of a gradient across the Gulf of Guinea and Sahel. The lower values (0.17–0.60) are recorded over the Gulf of Guinea and the highest (> 0.80) in the Sahel. This gradient explains that precipitation is concentrated in fewer months over the Sahel than over the coastal areas. The same gradient dynamics is observed in the case of future projections, although the PCD values are reduced, compared to the historical period. The lower values here are between 0.17 and 0.50, and the higher values are between 0.5 and 0.6. For the future period, the precipitation concentration decreases, and the Savanna and Sahel have the same precipitation distribution. Furthermore, the GWLs indicate that the rainy season will start earlier than in the present (historical period). The highest concentration period for the Gulf of Guinea and the Savanna will be from May to July, while the concentration over the Sahel will be highest in August.
Daily Variability of Precipitation
Figure 11 presents the variations between the projection of each GWL and the present period. Higher values of CDD are observed in the northern part of the study area, while higher CWD values occur in the coastal areas. Comparing the patterns of Figs. 11a–d, i–l, it can be seen that the CDD decreases about 10 ± 5 days over the northeastern part of the study domain, both annually and during the rainy season. In the northern part, a significant variability of dry days occurs within the rainy season (for instance, a reduction over the northeastern area and an increase over the northwestern area), which means that the northeastern of the study domain is wetter under GWLs and the northwestern area is drier. The Gulf of Guinea has a very slight variability in respect of CDD for all GWLs studied, at both annual and rainy season scales. The projections of the GWLs 1.5 °C, 2.0 °C, 2.5 °C show essentially the same variability in the annual CDD, while the GWL 3.0 °C (Fig. 11i) shows a significant increase in the annual CDD. The CDD is projected to increase for 4–5 days over the Gulf of Guinea; in Mauritania and Senegal, the increase is projected to be 10 ± 2 days. Niger and Chad (which are characterized by a dry northeasterly flow crossing the Sahara desert) are projected to record a reduction of CDD with a range of 12 ± 2 days. This agrees with the results of Klutse et al. (2018), who illustrated a decrease for GWLs 1.5 °C and 2.0 °C, in terms of the number of CDD in West Africa during the rainy season, and the results of Sultan and Gaetani (2016), who reported a reduction in the number of dry days over central Africa.
In general, the CWD does not appear to record as many variations as was the case with the CDD. It varies slightly with 0 ± 3 days. Nonetheless, high and important variations can be noticed at several specific points. Figure 11e–h shows that CWD is projected to decrease by 10 ± 2 over the southern parts of Benin and Nigeria. A small increase in CWD of up to 2 days is also likely to be recorded over the Sahel.