Climate change impact on Northwestern African offshore wind energy resources

The ROM model (REMO-OASIS-MPIOM) was used to produced four simulations, in stand-alone atmosphere (ROM_U) and in atmosphere–ocean coupled modes (ROM_C), for two time slices: a historical climate run encompassing the 1976–2005 period, and a future climate simulation spanning the period between 2070–2099 under the RCP8.5 scenario (Riahi et al 2011). Lateral boundary conditions were taken from a CMIP5 simulation with the Max-Planck Institute Earth System Model (Giorgetta et al 2013). The ROM domain, common for all simulations, covers the African continent, the Mediterranean region, and a large part of the Atlantic and Indian Ocean (figure 1(a)). The atmospheric resolution is 0.22°, with 31 hybrid vertical levels and three-hourly output. The vertical levels information allows us to feature the wind at upper atmospheric levels by interpolation.

Zoom In Zoom Out Reset image size

Additionally, 19 RCM simulations from the Coordinated Regional Climate Downscaling experiment (Giorgi et al 2009) over Africa (CORDEX-Africa) (Hewitson et al 2012), with a horizontal resolution of 0.44°, are used in this study (table 1). The 10 m daily mean wind speeds are considered for the historical climate (1976–2005) and for the RCP4.5 and RCP8.5 future climate scenarios (Moss et al 2010, van Vuuren et al 2011), for the mid-21st century (2040–2069) and end-21st century (2070–2099). Within the first hundred meters of the atmosphere, only this data is available. The unavailability of vertical level information requires the use of an extrapolation method to feature the wind at upper levels, as explained further.

Soares et al (2019b) performed a qualitative and quantitative evaluation of the near-surface wind speed from the uncoupled and coupled ROM simulations, and from the CORDEX-Africa RCMs, comparing their results against the Cross-Calibrated Multi-Platform (CCMP) dataset (Atlas et al 2011). In Soares et al (2019b), in order to assess the performance of the different RCMs, a set of statistical metrics were computed for each grid point and time scale (monthly, seasonal, and yearly) such as, bias, mean absolute error, root mean square error, normalized standard deviation, spatial correlation, and the Willmott-D score. These metrics allowed the analysis of mean values at the different timescales. Additionally, to measure the differences between distributions, analysis of the probability density function (PDF) between model and observations was done through a PDF matching score and the Yule–Kendall skewness measure.

To perform a robust assessment of the projected changes of the wind resource, a CORDEX-Africa multi-model ensemble was built considering the relative performance of each CORDEX-Africa RCM. All the above-mentioned metrics were included in the ensemble-building process, allowing the ranking of the models and the attribution of weights for the multi-model ensemble building (a detailed description of the evaluation process and ensemble building can be found in Soares et al (2019b)). This CORDEX-Africa multi-model ensemble (MME) is used in the current study to obtain a more robust description of the regional surface atmospheric flow and its climate change signal. The validation of the near-surface wind showed that ROM_C presents the best performance in reproducing the surface wind in the present climate, and that the CORDEX-Africa MME outperforms most of the individual RCMs, giving confidence for us to perform the climate change assessment in this paper.

In agreement with Pryor and Barthelmie (2011), the instantaneous wind energy density E can be computed as the power per square meter (P/A), and is given by

$$\begin{eqnarray}&&E=\displaystyle \frac{P}{A}=\displaystyle \frac{1}{2}\rho {v}_{z}{}^{3}\end{eqnarray} \tag{ 1 }$$

where ${v}_{z}$ is the wind speed at a height z and ρ is the air density, here considered constant and equal to $1.2\,{{\rm{kgm}}}^{-3}$ according to a standard atmosphere from the International Organization for Standardization. In general, the wind turbines do not produce electrical power for wind speeds below 3 ms⁻¹ and above 25 ms⁻¹. To characterize the present and future climates of the offshore wind resource, the wind energy density is computed at 100 and 250 m heights for wind speeds between 3 and 25 ms⁻¹, but the rated regime is not considered (ignoring the rated regime implies that the results can be strongly impacted by individual windy events). Since the wind speeds at 100 and 250 m are not available for any of the models, an interpolation or extrapolation of this variable is needed. For ROM simulations, the wind speed is interpolated from the model-levels output. For CORDEX-Africa simulations, the 100 and 250 m wind speed is calculated from the wind speed at 10 m height, using the logarithm wind profile extrapolation (equation (2); Yamada and Mellor 1975):

$$\begin{eqnarray}&&{v}_{z}={v}_{{z}_{m}}\,\mathrm{ln}\left(\displaystyle \frac{z}{{z}_{0}}\right)/\,\mathrm{ln}\left(\displaystyle \frac{{z}_{m}}{{z}_{0}}\right)\end{eqnarray} \tag{ 2 }$$

Here, ${v}_{{z}_{m}}$ corresponds to the 10 m wind speed and the ${z}_{0}$ is the local roughness length with a constant value of $1.52\times {10}^{-4}\,{\rm{m}}$ over the ocean surface (Carvalho et al 2014).

The wind energy density is calculated at a daily scale since it is the time scale available in all ROM and CORDEX-Africa models. However, for both ROM simulations, it is also calculated at a sub-daily scale (three-hourly).

At the annual and seasonal scales, the present climate wind energy density at 100 m height, given by the two ROM runs (uncoupled and coupled) and by the CORDEX-Africa MME are in general similar (figure 2). The annual energy density pattern and its seasonal cycle clearly show the imprint of the persistent northerly alongshore flow (Lima et al 2018, Soares et al 2019a), where the larger amount of available resource is present in the coastal regions, downstream of Cape Ghir and between Cape Bojador and Cape Blanc. The annual energy density at those locations can amount up to 800 Wm⁻², and then decrease to around 400 Wm⁻² in offshore areas and less than 200 Wm⁻² in the southern regions. The ROM uncoupled and coupled results of annual wind energy are quite similar, but the CORDEX-Africa MME presents larger values in the coastal regions offshore Western Sahara.

Zoom In Zoom Out Reset image size

The wind energy density seasonal cycle in coastal regions is sharp and closely linked to the CLLJ seasonal cycle. In winter, the CLLJ is almost completely absent, increasing its frequency of occurrence in spring and further in summer, when it reaches values above 70% of the time, and decreases in autumn to ∼40% (Soares et al 2019a). Correspondingly, the winter wind energy density in the coastal regions of Western Sahara shows values bellow 400 Wm⁻², increasing to 900 Wm⁻² in spring and reaching values up to 1400 Wm⁻² during summer. In autumn, the wind energy density is reduced to 600 Wm⁻² for the ROM runs and 800 Wm⁻² for the CORDEX-Africa MME. The seasonal maximum values occur in summer, downstream of Cape Ghir, where the coastal jet is more persistent, and in the Canary Islands shading vortices.

The higher horizontal resolution of ROM simulations provides a better representation of the alongshore flow when compared with the CORDEX-Africa results. Furthermore, coupling atmosphere and ocean leads to an improved representation of the surface wind. It should be kept in mind that the surface wind speeds from the RCMs analyzed here are usually stronger than the CCMP winds, especially closer to the shore (Soares et al 2019b). Despite this, these authors highlight the good agreement with observations of the ROM runs and of the CORDEX-Africa MME. Importantly, it was also mentioned that there are known problems of the CCMP in representing the near-surface wind close to shore due to pixel land contamination (Tang et al 2004). Another important issue refers to the impact of the logarithm extrapolation. Figure S1 (supplementary material), available online at stacks.iop.org/ERL/14/124065/mmedia, shows the wind energy given by the ROM simulations but extrapolating the 10 m wind speed to 100 m height, which display similar wind energy values with the CORDEX-Africa MME. In general, the logarithmic extrapolation results in an overestimation of wind energy density in broader areas, except to the conspicuous offshore areas and seasons where the CLLJ is more intense, such as Cape Ghir in summer. Finally, the impact of computing the wind energy at a daily scale is addressed in figure S2 through the relative difference between the computation of the wind energy density at a sub-daily scale (three hourly) and at a daily scale. There is an increase of the available wind energy density over all the domains, which are larger closer to the coast. The differences do not exceed 9% at the annual scale and 16% at the seasonal, but these occur in very limited areas where the wind energy density shows large values. For example, at the annual scale, the energy density in those areas can reach values of 800 ± 120 Wm⁻² downstream of Cape Ghir, and 400 ± 70 Wm⁻² between Cape Bojador and Cape Blanc; at the seasonal cycle, the higher values are found in coastal regions of Western Sahara during summer (1400 ± 150 Wm⁻²). This issue reveals the significance of the model's higher output sampling for wind energy assessment.

The future projections for the 100 m offshore wind density, from the RCP4.5 and RCP8.5 scenarios, and their changes, compared with the historical climate, are portrayed in figures 3 and 4. A Student's t statistical significance test was applied for all the changes projected by the ROM simulations, CORDEX-Africa MME, and individual CORDEX-Africa models (shaded areas are non-statistically significant at a 90% confidence level). To better understand the regional climate change signal and its coherence among the different simulations, three areas (B1, B2, and B3) were defined (figure 1(c)). The selected areas are linked to the regions of maxima NACLLJ frequency of occurrence and where larger gains of wind resource are projected (areas B1 and B2), and the region where significant future reductions are foreseen (area B3). The figure S3 shows the seasonal relative changes for the 100 m wind energy density between the RCP8.5 future climate (2070–2099) and the historical climate (1976–2005) for each area indicated in figure 1(c).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The annual and seasonal future spatial wind energy patterns are unchanged for both climate changes scenarios, displaying the same locations of higher availability of energy (figure 3). Nevertheless, there are some changes in wind energy values (figure 4). Both ROM simulations and the CORDEX-Africa MME project a decrease of the annual wind energy in the southern offshore regions. This reduction is especially large in the case of the uncoupled ROM simulation, reaching −30% for the RCP8.5. The CORDEX-Africa MME and the coupled ROM run project a decrease of ∼−15% for the RCP8.5, and for the RCP4.5, the CORDEX-Africa MME points to ∼−8%. Also, in coastal regions of Morocco, both ROM simulations project an increase in annual wind energy density up to +12%. Seasonally, the future wind energy patterns reveal a larger degree of change, especially in the ROM simulations. All models (ROM and CORDEX-Africa MME) project winter increases of wind energy for the offshore areas of Western Sahara, also for spring offshore Cape Ghir (up to +16% in winter and +30% in spring, for the ROM runs). Common to all simulations is a relevant reduction in the wind energy in southern areas for all seasons. The ROM uncoupled run points to a decrease higher than −30% in large southern areas offshore of Guineas in summer and autumn, although the ROM coupled and the CORDEX-Africa MME reveal smaller reductions that may still reach ∼−20 to −25%. Finally, the ROM simulations also point to relevant increases of wind energy in summer and autumn, yet confined to smaller offshore areas, offshore Senegal and Western Sahara, respectively.

Regarding the individual performance of the CORDEX-Africa models (figure S3), at least 2/3 of the CORDEX-Africa individual models agree as to the projected signal in all areas and the changes are statistically significant, giving likelihood and robustness to the CORDEX-Africa MME projections. In area B2, the 100 m wind energy density winter changes projected by the individual models reveal high agreement in the projected signal, resulting also in a robust result. Here, 68% of the individual models agree as to the positive signal projected for spring season (likely change). In the southern area, 90% of the individual models agree as to the projected decrease of the wind energy (very likely change), decreasing this likelihood to 73%, 68%, and 63% in spring, autumn, and summer, respectively.

The future relative changes of the energy density computed at a sub-daily scale (figure S4) reveal rather similar values to the ones at a daily scale, therefore ensuring consistency to the presented results. Furthermore, the sharper future changes given by the ROM runs, when compared with the CORDEX-Africa MME, are out of the scope of this paper but may be due to a number of reasons: ROM GCM forcing, the higher resolution used, the smoothing effect of the MME averaging procedure, etc.

For the RCP4.5 scenario, CORDEX-Africa MME projects minor changes for the seasonal wind energy, with statistically significant increases of up to +8% in offshore areas of Western Sahara in winter and in the vicinity of Cape Ghir in spring. The projected decrease in the southern regions include a vast region southward of Senegal occurrings in all seasons, reaching maxima reductions of ∼−20% in Guinea in autumn.

As expected, when looking at the wind energy density at 250 m height (annual and seasonal) in the future climate (figure S5 for RCP8.5), much larger values are projected to arise when compared with the ones at 100 m height. Annual wind energy is expected to reach values above 1000 Wm⁻² along the NACLLJ region. In accordance with the future increase of NACLLJ frequency of occurrence in spring and summer (Soares et al 2019b), the wind energy is projected to peak up to 1700 Wm⁻² in a significant stretch of coastal areas in the area of study. Accordingly, the wind energy relative future changes at 250 m (figure S6) are larger than at 100 m. For the ROM runs, this may amount up to +30% in spring in the vicinities of Cape Ghir, and values above +15% cover wide offshore regions.

Figure 5 depict the boxplots of the relative changes (RCP4.5 and RCP8.5 scenarios) of wind energy for 2040–2069 and 2070–2099, compared with to the present climate for each area. For the area B1, the annual wind resource is projected to slightly increase, (under ∼+7%) throughout the 21st century due to a strong spring increase of wind energy, which may reach values of ∼+20% for the ROM runs and values below ∼+10% for the CORDEX-Africa MME. This is in line with Soares et al (2019), who showed that during spring, the surface wind speed and frequency of NACLLJ occurrence is projected to increase. In winter and autumn, the wind energy is projected to be reduced in the B1 area.

Zoom In Zoom Out Reset image size

Similarly, the annual wind energy is projected to increase in area B2 by the ROM runs (<10%), and remain practically unchanged throughout the 21st century by the CORDEX-Africa MME. However, the seasonal cycle of relative changes is quite different. The projections of future increasing NACLLJ frequency of occurrence by the ROM runs (Soares et al 2019b) in all seasons impact positively the future wind energy in area B2. Particularly in spring and autumn, the wind energy is projected to increase in the range of +10 to +20%, in close relation with the temporal expand of the NACLLJ annual cycle. The CORDEX-Africa MME is overall in line with these changes, although with much smaller projected change values. Finally, for the area B3, the results show a relevant loss of wind energy resource at the annual and seasonal scales. The projected decreases are more severe across winter and autumn, when the seasonal values are projected to decrease ∼−30% by the ROM simulations and ∼−15% by the CORDEX-Africa MME.

Based on the occurrence of the NACLLJ results from Soares et al (2019b) for area B1, the relationship between the energy density at 100 and 250 m heights was computed by model level interpolation and using the logarithmic extrapolation (figure 6), for present and future periods, when NACLLJ occur and when there is no NACLLJ occurrence. The black and red dots represent the maximum value of the daily mean energy density of the area when there is no jet occurrence at all times of the day (eight times) and when there is jet at all times of the day, respectively (computed by model-level interpolation). The blue and green dots refer to the daily mean wind energy density for the same grid point, but are computed by logarithmic extrapolation. These results are given by the coupled ROM run. For all seasons, with or without NACLLJ occurrence, it is clear from figure 6 that the logarithm extrapolation leads to a general underestimation of the relationship between the wind energy density at 100 and 250 m. In spring, the increase in the frequency of the coastal jet (in area B1) is clear and determines higher values of wind energy density, particularly at 250 m, linked to the median height of NACLLJ occurrence of ∼350 m (Soares et al 2019b).

Zoom In Zoom Out Reset image size