Projected changes in wind speed and wind energy potential over West Africa in CMIP6 models

Output from 14 CMIP6 models (Eyring et al 2016; see table 1 for institutions, model names, primary reference, and spatial resolution information of each model) simulations that was available at the time of this analysis were used in this study. For each model, daily 10 m surface wind speed and directions from the first realization ('r1i1p1f1') for both the historical and projected (shared socioeconomic pathways; SSP5-8.5; O'Neill et al 2016) experiments were used. Since actual ground-based wind speed and direction dataset over West Africa are sparse and not publicly available, we used the recently released ERA5 (Hersbach et al 2019) reanalysis near surface wind speed and directions for model's validation. The ERA5, a fifth-generation reanalysis developed by the European Centre for Medium-Range Weather Forecasts has a horizontal grid spacing of about 31 km. This version has higher spatial and temporal resolution and various improvements.

The CMIP6 models and ERA5 datasets were regridded to a common grid of 2.81° × 2.81° (latitude × longitude) using an area-conserving remapping procedure, which is implemented in the Climate Data Operators (https://code.zmaw.de/projects/cdo) to produce multi-model summary statistics based on the lowest model resolution. In order to reduce uncertainty in this study, the multi-model ensemble mean of all the CMIP6 simulations defined herein 'EnsMean' was used following Akinsanola and Zhou (2019).

The WPD (unit: W m⁻²) is an important measure for assessing the potential of wind energy generation in a location (Emeis 2013), and is defined in equation (1) as

$$\begin{equation}{\text{WPD}} = \,\frac{1}{2}\,\rho {V^{\,3}}\end{equation} \tag{ 1 }$$

where V is the wind speed at the adjusted-to-turbine hub height (taking herein as 90 m), and $\rho$ is the air density (standard conditions are assumed for $\rho$ with a constant value of 1.225 kg m⁻³). In order to compute the WPD, most studies extrapolate the 10 m surface wind speed to the desired turbine hub height. There are numerous methods of extrapolating 10 m surface wind measurements to the turbine hub height. The power law method (Emeis 2005) defined in equation (2), used in this study assumes that wind speed at a certain height z is approximated by

$$\begin{equation}{\text{wspd}}\left( z \right) = {\text{wspd }}({z_{{\text{ref}}}})\,{\left(\frac{z}{{{z_{{\text{ref}}}}}}\right)^{\alpha}}\end{equation} \tag{ 2 }$$

where z is the height at turbine hub height (taking as 90 m), ${z_{{\text{ref}}}}{ }$ is the reference height, wspd(${z_{{\text{ref}}}}$) is the wind speed at ${z_{{\text{ref}}}}$, and $\alpha$ is the power law exponent. The near-surface wind speeds at 10 m is used in this study as the reference height wind speeds and we assume that $\alpha$= 1/7, as it is appropriate for open land surfaces and has been used in previous studies over the region (e.g. Akinsanola et al 2017, Sawadogo et al 2019). In this study, the WPD was calculated at each grid point over the study area for both the CMIP6 simulations (historical and SSP5-8.5) and ERA5 dataset. All the analyses and calculations presented in this study are for annual and entire seasonal means defined herein as: ANN (January–December), DJF (December–January–February), MAM (March–April–May), JJA (June–July–August), and SON (September–October–November), and were integrated over the West African domain (see figure S1 (available online at stacks.iop.org/ERL/16/044033/mmedia) in the supplementary information) and further assessed over the three subregions namely: Guinea coast (4°–8° N, 20° W–20° E), Savannah (8°–11° N, 20° W–20° E), and Sahel (11°–16° N, 20° W–20° E). The ability of the CMIP6 models to reproduce the regions near-surface wind speed is first evaluated by comparing the historical simulations with ERA5 reanalysis. Both spatial assessment and statistical approach (percentage bias, normalized root mean square error (NRMSE), and the pattern correlation coefficient (PCC)) were used to evaluate these models. We compute the projected changes by comparing two 30 year time slices from the projections (2040–2069, 2070–2099) to the historical period of 1985–2014, and the statistical significance was evaluated using a t-test. Also, grid points where at least 70% of the ensemble members agree on the sign of change in the EnsMean were further assessed, which provides further insight into the robustness of the projected changes.

The spatial distributions of annual and seasonal mean wind speed and direction is presented in figure 1 for ERA5 reanalysis, EnsMean, and their relative biases (i.e. EnsMean minus ERA5) during the historical period of 1985–2014. The EnsMean is able to skillfully reproduce the spatial wind distribution (both the speed and direction) observed in the ERA5 on both annual and seasonal timescales. Broadly, the average offshore wind speed is nearly twice the onshore wind while wind direction was majorly south-easterly (north-easterly) below (above) the equator with a range of 2–8 ms⁻¹. Near the shore and onshore, the wind direction is dominated by north-easterly in DJF (figure 1(d)), northerly in MAM (figure 1(g)), south-westerly in JJA (figure 1(j)), and partly north-easterly and easterly in SON (figure 1(m)) above the equator. Over the continental area (i.e. offshore), the wind speed is maximum in DJF onshore around the Sahelian region with a magnitude of about 8 ms⁻¹, reducing through MAM, JJA, and SON to about 5 ms⁻¹. The spatial plots show that coastal areas in Liberia, Sierra Leone, Guinea, and Gabon recorded the lowest wind speeds of about 2–4 ms⁻¹ on both annual and seasonal timescales.

Zoom In Zoom Out Reset image size

The EnsMean (figures 1(b), (e), (h), (k) and (n)) was able to capture most of the climatology details observed in ERA5 reanalysis similar to the findings of Sawadogo et al (2020), although considerable biases still exist. For instance, for the annual spatial wind speed (figure 1(c)), the EnsMean significantly overestimated the ERA5 over the Sahel and western half of Nigeria by about 30%–40%. Below 5° N, the EnsMean underestimate the annual wind speed by about 10%–20%, evident over southern Nigeria and Cameroun. These biases are statistically significant at 95% level. Seasonally, underestimation by the EnsMean was similarly more pronounced below 5° N and east of 8° E (around Cameroun) with peak value of about −40% observed in DJF and JJA and statistically significant only in the latter. Wind speed was overestimated by about 40%–60% in northern Nigeria albeit significant only in the north-east in DJF, while underestimations of only about −10% to −20% is observed below the 10° N that is significant only in parts of Guinea and Sierra Leone, as well as Ghana. MAM (figure 1(i)) showed slight levels of overestimation above the 10° N and in Sierra Leone while the region below 5° N latitude is again under-estimated. Underestimation (overestimation) is significant and high (i.e. about 30%–50%) around south-western Cameroun and Equatorial Guinea (Nigeria, Benin, Burkina Faso and Mali) in JJA shown in figure 1(l) while SON (figure 1(o)) appeared to have the least average biases as only a few areas shows high large differences with majority of the study domain showing a significant overestimation (underestimation) of about 10% above (below) the 6° N latitude.

The ability of the ensemble members alongside their EnsMean to realistically represent historical 10 m wind speed is further assess across all the three subregions (i.e. Guinea coast, Savannah and Sahel) using a portrait diagrams and results presented in figure 2. Relative to ERA5, the percentage bias, NRMSE, and pattern correlation of the CMIP6 wind speed was investigated. Most of the models including the EnsMean showed percentage bias within the ±20% range over the Guinea coast although the negative bias was more than the positive bias. MRI-ESM2-0, MPI-ESM1-2-LR, MPI-ESM1-2-HR, MIROC6, and IPSL-CM6A-LR (BCC-CSM2-MR) all showed negative (positive) bias for the ANN, DJF, MAM, JJA, SON timescales as shown in figure 2(a) indicating an underestimation (overestimation) of near surface wind speed. INM-CM5-0, INM-CM4-8, and GFDL-ESM4 showed zero bias predominantly but with some overestimation (i.e. positive bias) in MAM and DJF. The Savannah subregion (figure 2(b)) showed a lot of positive biases, peaked around 40%–50% in BCC-CSM2-MR model for all timescales. Sahel on the contrary (figure 2(c)) exhibited varying levels of positive biases in most models and more pronounced in MRI-ESM2-0 and BCC-CSM2-MR models. One major difference between the Sahel subregion and others (i.e. Guinea coast and Savannah) is that the EnsMean also showed considerable positive bias, perhaps due to the individual model performances over the subregion. The bias at annual timescale appeared to be minimal over the three subregions when compared to the seasonal statistics. Overall, the CMIP6 multi-model mean performs better than most individual models at capturing the near surface wind speed over the subregions.

Zoom In Zoom Out Reset image size

Figures 2(d)–(f) shows the results of the NRMSE scaled from 0 to 1. Over the Guinea coast (figure 2(d)), only MIROC6 exceeded 0.6 in DJF while MIROC6, IPSL-CM6A-LR, INM-CM5-0, INM-CM4-8, and EnsMean all have NRMSE of about 0.4 across the timescales. In the Savannah subregion (figure 2(e)), IPSL-CM6A-LR had the highest NRMSE of 0.8 followed by NRMSE of 0.6 in BCC-CSM2-MR in JJA and SON, and MIROC6 in DJF. Compared to the Guinea coast subregion, most of the models had NRMSE of around 0.2 for all timescales considered in the Savannah subregion. Most of the CMIP6 models in the Sahel subregion (shown in figure 2(f)) exhibit smaller NRMSE value (less than 0.25) across all timescales. The EnsMean performance is relatively similar across all the timescales and subregions. Furthermore, the PCC results for the models are shown in figures 2(g)–(i) for the Guinea coast, Savannah, and Sahel subregions respectively. For all the timescales considered, all the CMIP6 models had a PCC greater than 0.9 essentially indicating a good level of agreement with the ERA5. Peak PCC value of 0.98 was observed at least once in all the subregions and all the seasons albeit most pronounced in the Sahel subregion at annual timescale. These results provide reliable information on individual model performance in reproducing near surface wind speed over the region, addressing a research gap identified by Sawadogo et al (2019). Most of the CMIP6 models alongside their EnsMean exhibited relatively high PCC, low percentage biases and NRMSE thus highlighting a realistic representation of near surface wind speed over West Africa. Also, our results demonstrate that no single model is consistently the most reliable across all the timescales and subregions.

The spatiotemporal projected changes in wind speed at turbine hub height (i.e. 90 m) for the period 2040–2069 and 2070–2099 relative to the historical period of 1985–2014 is presented in figure 3 (percentage values) and figure S2 in the supplementary information (absolute values) for annual and seasonal timescales. At the annual timescale, the EnsMean project a statistically significant increase in hub-height wind speed for both 2040–2069 and 2070–2099 period by more than 40% along the 4°–11° N latitude belt, this increase is robust as at least 70% of the ensemble members agree on the sign change in the EnsMean. We found a robust statistically significant slight reduction (around −8%) in the eastern Sahel, evident only in the 2040–2069 period while parts of western Sahel exhibit similar robust decrease in 2070–2099 period. The observed projected changes at annual timescale is similar for all the seasons, although most grid points exhibit projected increases in wind speed than decreases and intensified towards the end of the century (2070–2099) than mid-century (2040–2069).

Zoom In Zoom Out Reset image size

More specifically, in DJF season (figures 3(c) and (d)), there is a projected robust reduction in wind speed over the entire Sahelian belt. These future decreases are statistically significant over most grid point and at least 70% of the model agree on the sign of change of the EnsMean. Future intensification dominates major parts of the Guinea coast, although only few grid points are statistically significant and have at least 70% intermodel agreement. The region of robust increase includes parts of the Fouta Djallon highlands, Ghana, and eastern Cameroon. The overall changes are consistent between the mid and end of 21st century, although relative spatial differences exist. In MAM shown in figures 3(e) and (f) however, there are slight differences compared to DJF and ANN both spatially and temporally. Strong increases that are statistically significant are observed in the Guinea coast and southern Savannah, this reduces eastward from about 16% to about 4%. Northwest of 11° N, there is a significant projected decrease in wind speed of around 6%. As in DJF, the projected late century MAM wind speed (figure 3(f)) is considerably higher (lower) for areas with projected increase (decrease) when compared to the mid-century albeit with similar spatial changes.

Furthermore, the JJA and SON (figures 3(g)–(j)) seasons show the highest projected robust changes. The projected changes in most grid points across West African have either statistically significant or have at least 70% of the ensemble members agreeing on the sign of change in the EnsMean. The JJA exhibit widespread spatial future intensification in wind speed that range from 14%–18% (2%–6%) below (above) 10° N in 2040–2069 and about 16%–20% (4%–8%) also observed below (above) the 14° N in 2070–2099. In terms of the absolute values (figure S2 in the supplementary information), the sign of the projected changes are spatially consistent across all the timescales with the percentage values. Robust projected increase is also evident in JJA and ranges from +0.2 to +0.6 m s⁻¹ (+0.6 to +1.2 m s⁻¹) in 2040–2069 (2070–2099). Worthy of note is a slight statistically significant increase (although with no intermodel agreement) in wind speed for both mid-century and late century over the eastern Guinea coast. In SON (figures 3(i) and (j)), there is a more widespread future decrease in wind speed above the 10° N that is statistically significant east of the 2° W alongside high intermodel agreement. The magnitude of the decrease is about 4%–10% in the 2040–2069 and 6%–16% in 2070–2099 period. Below 10° N, projected intensification in wind speed reaching around 16%–20% dominates the mid-century period. Similar but intense increase is evident in the late century. The overall changes are robust considering the high intermodel agreement and widespread statistically significant grid points.

To provide more robust study results, as well as accommodate the cubic relationship between wind speed and wind power, the projected changes in WPD based on the EnsMean is examined and presented in figures 4(a)–(j) for percentage values (figures S3(a)–(j) in the supplementary information for the absolute values). The WPD is projected to increase annually by about 50%–60% in most parts of the Guinea coast subregion in the mid-century (figure 4(a)), specifically in parts of Sierra Leone, Liberia, Cote D'Ivoire, Ghana, Togo and Benin. Eastward intensification towards Nigeria and Cameroun is evident in the late century (figure 4(b)). The observed increases are statistically significant and have majority of the ensemble members agreeing on the sign of change in the EnsMean. Although slight statistically significant projected decrease (about 15%) is evident around 12°–20° E, majority of the models disagree on the sign of change in the EnsMean in the late century.

Zoom In Zoom Out Reset image size

In figure 4(c) along the Savannah and Sahel (Guinea) subregions, there is statistically significant projected WPD decrease of about 10%–20% (20%–30%) in the west (east) of the study domain during DJF. On the west of the Guinea coast however, there is some increase in WPD with a range of 30%–50% significant only in the coastal parts of Sierra Leone, Liberia, Cote D'ivoire, Ghana, Togo, Benin, and Nigeria; and in some countries below the 5° N latitude. For the late century i.e. 2070–2099 (figure 4(d)), a statistically significant slight decrease of WPD is observed over the 12°–18° N belt that shifted southward to the 10°–15° N belt east of the 6° E longitude. In MAM, the spatial characteristics for both mid-century (figure 4(e)) and late century (figure 4(f)) are similar but the magnitude is higher in the latter. Most grid points over the eastern (mid-) Guinea coast exhibit projected increase in WPD of about 30%–60% (40%–60%) that is both statistically significant and has majority GCMs agreement while the areas above the 10° N appear to have less than 10% decrease. The magnitude of the increase is higher in 2070–2099 relative to 2040–2069 especially along the mid-Guinea coast gaining almost 10%–20% more WPD.

The projected WPD increase observed in JJA (figures 4(g) and (h)) is widespread, albeit higher in the Guinea coast and Savannah than Sahel. For the 2040–2069 period, an increase of about 60%–80% is observed around and below the 11° N latitude while for the 2070–2099 period, the maxima line advances northward to the 15° N. Notably also, the increase in WPD in the north of the study domain is more pronounced and stronger in the late century than mid-century. The increase in all these areas and regions were statistically significant, and majority of the individual models agree with the observed increase in the EnsMean. Similar patterns are observed in SON (figures 4(i) and (j)), although the changes are more diverse with the areas above (below) the 10° N showing future decrease (increase) in WPD of about 10%–30% (40%–70%) in the mid-century. In both mid-century and late century, all areas with significant increase (decrease) in WPD have high intermodel agreement. The results of the absolute projected changes presented in figure S3 in the supplementary information are spatially consistent across all the timescales with the relative changes.

Figures 5(a)–(f) shows the projected changes in WPD for all the CMIP6 ensemble members alongside their EnsMean across all the three homogenous subregions of West African and all the timescales, for both the mid-century and late century. The overall changes for the ensemble members range for 2040–2069 (2070–2099) period are −22% to +50% (−25% to +65%), −28% to +60% (−40% to +80%), and −40% to +50% (−55% to +95%) for Guinea coast, Savannah, and Sahel subregions respectively. For the Guinea coast subregion (figures 5(a) and (b)), across all the timescales considered, the EnsMean exhibit a projected increase in WPD except DJF in the 2070–2099 period. The EnsMean projected WPD increased by about 10%–28% over the subregion, with high intermodel agreement among the ensemble members, except in DJF in both mid-century (figure 5(a)) and late century (figure 5(b)) where model agreement was split 50:50. The projected increase was maximum in the JJA season and least in DJF for 2040–2099, while it was maximum in JJA and least in ANN in 2070–2099 period. Under the conventional RCP8.5 scenario, Sawadogo et al (2019) similarly projected an increase in WPD over the Guinea coast subregion dependent on warming level. In the Savannah subregion shown in figures 5(c) and (d), DJF again shows a decrease in WPD for both mid and late century. Projected changes intensified in 2070–2099 from mid-century values, supported by high intermodel agreement. DJF showed a decrease in WPD for both mid and late century with 100% intermodel agreement, but ANN and SON despite projected increase had only about 70%–80% of the models agreeing to the projected increase. The EnsMean projected increase is about 7% (10%) in ANN, 4% (10%) in MAM, 22% (34%) in JJA, and 12% (21%) in SON for mid (late)-century. At least three of the models projected WPD decrease in ANN and SON for mid-century (figure 4(c)) and late century (figure 4(d)), creating a range of −30% to +45% for 2040–2069 and −30% to +62% for 2070–2099 in SON. Furthermore, the WPD in the Sahel subregion is projected to decrease in all seasons except JJA for both mid and late century as shown in figures 5(e) and (f) respectively. WPD increased by about 12% (28%) in mid (late)-century, with all but three models (21%) agreeing on the sign of change over a range of −5% to +50% (−5% to +95%). At least ten of the models (71%) agree with the decrease over the subregion typified in the EnsMean for ANN, DJF, MAM, and SON while the remaining three models (21%) projected increase in WPD over the subregion, especially in ANN and SON where one model projected about 10% increase for both mid-century and late century. Across all the time scales and subregions, the sign of the absolute projected changes presented in figure S4 in the supplementary information is consistent with the percentage values.

Zoom In Zoom Out Reset image size