1. Introduction
The critical role of the ocean in the Earth’s climate system is well known and documented, as the ocean is the largest carbon sink and heat reservoir absorbing over 93% of heat generated since the Industrial Revolution [
1]. However, evidence in the scientific literature affirms that the last few decades have witnessed a significant increase in the heat storage capacity of the ocean with a deleterious impact on key components of the Earth–atmosphere system [
2]). The observed ocean-warming manifestation includes an increase in the average sea surface temperature and changes in ocean circulation, stratification, heat transport, ocean biogeochemistry, and marine heat waves with concomitant effects on regional and global climate variabilities and human societies [
3]. It is on record that the average global sea surface temperature (SST) of the ocean has had a significant warming trend of ~0.13 °C per decade since the beginning of the 20th century, with the SSTs of the last three decades considered the warmest since the commencement of modern instrumental records [
4] with global warming implicated in the observed warming effect.
Several studies have been conducted over the years to examine global sea surface temperature distributions and their influence on climate extremes over the West African region. The above assertion hinges on the fact that the amount of near-surface moisture for convective rainfall is sensitive to sea surface temperature anomalies through the Clausius–Clapeyron relationship [
5]. Another mechanism where SST influence is evident is in the tropical Atlantic Ocean through the interaction with the wind field, in which the weakness of the trade winds over the Equator forces the Intertropical Convergence Zone to migrate northwards, thereby strengthening southerly trade winds. This interaction ultimately leads to upwelling, vertical missing, and evaporation into the atmosphere, and eventually leads to a phenomenon called cold tongue, which develops in May, June, and July, persisting till September. It is this seasonal SST anomaly that exercises significant control over rainfall patterns in the West African climate [
6].
Consequently, modeling studies by [
7,
8] corroborate the above analogy on the influence of ocean forcing on rainfall variability in Sub-Saharan Africa. Similar studies have also focused on the impacts of localized SST anomalies over the coast of Guinea and the eastern Atlantic Ocean [
9]. Ref. [
10] studies in 1987 were more emphatic as they revealed that the reduction in rainfall in the Sahel during winter corresponded well with the warm SST anomaly of the Guinea coast; the global sea surface temperature (SST) variability merely slightly moderated its effects.
Ref. [
11] in his study assessed the impacts of warming/cooling of the Atlantic Sea surface temperature (SST) on the climate in West Africa using Version 4.4 of the Regional Climate Model (RegCM4.4A.) The study outcome revealed that the 1–2 K cooling and warming of the Atlantic SST will result in triple temperature and precipitation change structures over West Africa as well as an anomalous dipole of precipitation, resulting in opposite signs over the Sahel and the Gulf of Guinea.
Given the scale and vulnerability of West Africa, described as one of the most populated regions of the world, to extreme climate events, climate change is manifested in an increased frequency of floods and droughts, with devastating impacts on people’s livelihoods and the attainment of sustainable development goals [
12]. The vulnerability is further driven by a range of factors that include a weak adaptive capacity, dependence on productive resources that are climate-sensitive, Africa’s geographical location in the lower latitudes, and widespread poverty climate change, which is also disrupting the hydrological cycle, putting a strain on water resources, population displacement, and food insecurity [
13]. This has led the scientific community to recognize the need to mobilize additional resources to mitigate the projected impact of climate change on critical sectors of the economy in the sub-region and to build resilience. This is against the backdrop that a warmer climate holds more moisture, implying an increased frequency of rainfall events. Conversely, rising temperatures accelerate evaporation and can intensify drought episodes [
14].
The Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) (AR4) projected a warming rate of 0.5–2.5 K over the Atlantic Ocean for the twenty-first century [
15] based on all the Special Reports on Emissions Scenarios (SRESs), while the Fifth Assessment Report of the IPCC also projected a temperature change of 0.2–2.3 K over West Africa [
16] based on all the Representative Concentration Pathways (RCPs). In the same vein, the Sixth Assessment Report (AR6) of the IPCC 2022 affirms that the global mean SST has increased since the beginning of the 20th century by 0.88 °C, with significant warming projected in the coming decades.
The increase in the SST is reported to be driven by climate change at an oceanic scale, such as the Atlantic Equatorial Mode and El Niño Southern Oscillation (ENSO) [
17,
18]. Thus, the understanding of future climate change effects on SST variability and warming in the Gulf of Guinea has become pertinent to investigate. The General Circulation Model (GCM) outputs of the Coupled Model Intercomparison Project (CMIP) are an essential dataset for forecasting future climate trends. The main difference between the previous CMIP5 models and the current CMIP6 output is based on the set of future scenarios used to project climate evolution. The CMIP6 model was designed to bridge and improve the restrictions identified in the CMIP5 models with a focus on identifying systematic errors in simulations and improving the representation of land use changes in the climate (IPCC report 2013). Several new scenarios are used by CMIP6 called Shared Socioeconomic Pathways (SSPs), which are combined with previous CMIP5 scenarios of climate radiative forcing called Radiative Concentration Pathways (RCPs). The GCMs of Coupled Model Intercomparison Project 6 (CMIP6) are the most advanced tools currently available for climate studies and are better than previous projects. Moreover, the GCMs of Coupled Model Intercomparison Project 6 (CMIP6) are also the most advanced tools currently available for climate studies and for simulating past and future extreme climate events, given the scale of anthropogenic-induced global warming [
19,
20,
21].
However, GCM simulations cannot be directly used for climate impact studies because of their inability to give reliable information at a local scale [
22]. Furthermore, GCM outputs are often coarse in their temporal and spatial dimensions, resulting in systematic biases [
23]. Hence, it is critical to validate the abilities of these models before use for any climate projection and impact studies. Therefore, downscaling model outputs is necessary to improve the model resolution to match the resolution at a local scale. Downscaling is essentially the process where spatial data are represented by lower spacing and smaller temporal intervals [
24].
Bias correction is a technique employed for resolving high-resolution Global Circulation Model (GCM) and Regional Climate Model (RCM) outputs known to exhibit systematic biases [
25]. This is important, given the fact that General Circulation Models (GCMs) provide valuable information dealing with historical and future larger-scale climate trends; unfortunately, as pointed out above, their resolution is too coarse to investigate the localized impact effects of extreme climate events [
26,
27]. Furthermore, as stated by [
27], raw GCM outputs are characterized by a non-trivial degree of bias and the seeming limitation that often affects the ability of GCMs to reproduce extreme tails of climate variables [
28]. It has therefore become imperative that, before GCM/RCM outputs are applied for use in hydrological [
29,
30], agricultural, and climate risk assessments, they must be downscaled to a finer resolution and bias-corrected as far as the observed data are concerned [
31,
32]. Hence, bias correction can be described as a post-processing technique in climate data to achieve a more realistic and consistent representation of atmospheric processes at very high and fine spatial scales.
The evidence in the literature affirms that statistical and dynamic approaches are the most common methods used for the downscaling and bias correction of projections involving climate variables from GCMs. The statistical approach is based on the distribution and relationship between the observed and projected data for the historical period [
29,
33]. Secondly, in statistical downscaling, statistical relationships between coarse-scale climate variables and locally observed data are established, including integrating the effects of fine-scale predictors into downscaled data [
34]. The dynamical approach, on the other hand, is based on a regional climate model forced with the boundary conditions from coarse-resolution GCMs [
35,
36].
The process of downscaling also requires the conversion of the model output from a course to a finer resolution, especially when a Regional Climate Model (RCM) is forced with a GCM, resulting in finer-scale output where regional climate processes, topography, and orography are incorporated [
37]. Despite the advantages of dynamical downscaling, its limitation stems from the fact that it is computationally intensive and can introduce additional biases [
38,
39]. Statistical downscaling has been extensively deployed in many studies due to its efficiency and can be applied to a variety of climate variables in topographically challenged terrain [
36,
40].
However, downscaling is complemented by bias correction, a procedure in which the climate model output is adjusted such that its statistical properties (e.g., mean, variance, and potentially higher moments) resemble those of observations in a common climatological period [
27,
41]. This study adopts the nonparametric statistical transformation-based bias correction, which shows better skills in comparison to the parametric method in reducing biases from GCM outputs [
42]. The success and applicability of bias correction for the projection of future climate variables were proven by several studies conducted by [
30,
43,
44,
45], indicating the performance and acceptability of the bias correction method.
Dynamical downscaling and statistical downscaling have emerged as some of the best methods widely used for post-processing. The statistical approach has advantages over dynamical downscaling as it is a lot less resource intensive [
33], Additionally, the pros and cons of both approaches are that, while statistical approaches are based on the distribution and relationship between the observed and projected data for the historical period, dynamical downscaling, on the other hand, is based on a regional climate model forced with the boundary conditions from the coarse-resolution global circulation model [
46]. Given the gap identified in the literature on the use of GCM CMIP6 models for simulating the historical and future SSTs of the Gulf of Guinea, it has become critical to examine the skills of CMIP6 models in simulating the sea surface temperature over the Gulf of Guinea.
The importance of the analyses of the historical and predicted future sea surface temperatures (SSTs) of the Gulf of Guinea cannot be overemphasized as they provide a framework to better understand and assess the predictability of extreme climate events and their impacts under a warming climate scenario.
The motivation for this research originates basically from the need to accurately understand and address future climate impacts in Africa both at the regional and local scales and the imperative for climate data products to represent climate extremes accurately and be available at fine and temporal resolutions [
47]. This requires the deployment of bias correction techniques to rectify the problem associated with GCM simulations to generate reliable information at a local scale, which can be used for the formulation of climate adaptation strategies in Africa in the face of the increased frequency and severity of climate extremes.
The objectives of this study therefore are to evaluate the performance of 8 GCM-CMIP6 models in simulating the historical sea surface temperatures of the Gulf of Guinea in 1970–2014. Secondly, we used bias-corrected CMIP6 outputs to simulate future SST projections (2015–2100) and finally used statistical techniques to quantitatively validate the performance of the models.
4. Discussion
The spatial climatology of the historical data (1940–2014) and the CMIP6 simulations (2030–2100) were presented in the preceding section. It is important to underline that most of the models successfully reproduced the spatial patterns of SSTs over the GOG and the Western Sahel. However, the highest sea surface temperature (SST) value was produced by CAM-ESM (35.48 °C and 22.39 °C), indicating that the model overestimated the SST climatology, while the lowest values were for 2030 (29.38° and 19.71 °cC), implying the persistence of a warming bias in the Guinean cost and cold tongue along the Western Sahel. These results clearly show the influence of ENSO as the dominant driver of SST variability as evidenced by the warm bias and cold tongue in the Western flank of the Guinean coast. This suggests that all the models overestimated SST for the Guinean coast when juxtaposed with the observed dataset. The historical ERA5 data are used to validate the projected changes in the SST and, as shown in the results presented in the preceding section, the bias-corrected data are consistent with the observed data for the climatological mean period.
The findings from the CMIP6 model simulation further reveal that the SST is projected to increase by 1.03 °C between 2030 and 2040 in the Western GOG, with most of the warming projected to occur in 2090 and 2100. The projected warming bias demonstrated by the CMIP6 models was unsurprising as RCP 8.5 assumed higher GHG emissions leading to fossil fuel development [
20]. Similarly, the SST for the Guinean coast known for its high warming was projected to increase by 4.61 °C between 2030 and 2100, with the SST expected to be substantially higher in the far than the near period in the study domain, although the projected warming was not homogenous as substantial increases in the multimodal mean SST were expected more often for the Guinea coast than the Western Sahel. Similarly, [
66], in their studies of bias-corrected CMIP 3 and CMIP 5, affirmed that coupled models exhibited sizable biases in the mean position of the West African Monsoon. Furthermore, the outcome of their analysis revealed that most models contained a warm bias in the Equatorial Atlantic and a southward shift of the ITCZ in coupled models; this southward bias was also examined in other studies conducted by [
67,
68]. Other studies that have been carried out using the GCM CMIP simulation include those of [
69] who asserted that the Coupled Model Intercomparison Project Phase 5 (CMIP5) historical simulations underestimated SST seasonal cooling from April/May to August over the northern Gulf of Guinea, and suggested that this could have been driven by the incorrect seasonality of precipitation over the southern coastline of West Africa. Similar studies conducted by [
70] on the projected sea surface temperature over the Equatorial Pacific posited that the warm pool region was projected to experience enhanced warming along the Equatorial Pacific. Overall, the future SST of the GOG is projected to become warmer in the future and the magnitude of change depends on future GHG scenarios. The projected warming presented in this study can impact marine ecosystems, fish, and the local climate.
4.1. Statistical Evaluation of the CMIP6 Models’ Performance
This section examines the ability of the CMIP6 models to simulate the observed historical simulation. Several methods of validating the performance of the GCM-driven CMIP6 simulation exist in the literature, and they include the Nash–Sutcliffe efficiency (NSE), the Root Mean Square Error (RMSE) or the Root Mean Square Deviation (RMSD), the Mean Absolute Error (MAE), the Percentage Bias (PBIAS), and the Correlation Coefficient (Cr). These techniques have been employed by different researchers for the validation of the bias-corrected CMIP6 simulation [
45,
71,
72,
73]. However, in this study, the performance of the GCM-CMIP6 model in simulating the observed dataset was evaluated using the correlation coefficient. Consequently, the ability of each of the models used in the study to reproduce the spatial pattern of the SST over the GOG was evaluated/validated against the ERA5 reanalysis data in 1940–2014 using a statistical analysis.
A regression analysis is very useful when it comes to studying the relationship between variables. The regression analysis can identify the cause and effect of one variable regarding another variable. Variables are the main part of a regression analysis. There are dependent variables (or criterion variables) and independent variables (or predictor variables). In a multiple regression, the independent variables can also be added to the model to explain the cause and effect of dependent variables. Hence, dependent variables can be predicted by building better models using a multiple regression analysis.
An overview of the models used and their respective countries, spatial resolutions which is 10° × 10° including the period of coverage for both the historical and the simulated are presented in
Table 2. The table further highlights the emission scenarios and shared socio-economic pathways. The table is significant as it provide a background of the models used in the study.
Multiple Regression
Multiple regression models can be presented by the following equation:
where
Y is the CanESM (dependent variable),
X1,
X2,
X3,
X4, and
X5 (independent variables) are ACCESS, CAMS-CSM, CMCC-ESM, EC-Earth3, MCM-UA, and MPI-ESM, respectively.
β1,
β2,
β3,
β4, and
β5 are the model coefficients of the six independent variables.
β0 is a constant, while ε is the error.
The outcome of the model regression analysis conducted, as shown in
Table 2,
Figure 12a–f and
Figure 13, indicates that all the models, namely, ACCESS, CAMS-CSM. CAN-ESM, CMCC, MCM-UA, and MPI-ESM have r values of 0.10 for ACCES, as shown in
Figure 10A; 0.0185 for the CAMS-CSM model, shown in
Figure 10B; and 0.0681 for CAN-ESM, shown in
Figure 10C. Similarly, CMCC denoted in
Figure 10D has R-values of 0.0919 and 0.107 for MCM-UA, as shown in
Figure 10E, with 0.013 for MP-ESM in
Figure 10F. The weakest correlation was the CAMS-CSM model, with an R-value of 0.01, while the strongest correlation was MCM-UA, with 0.13, given the fact that the closer the r-value moved to +1, the stronger the correction between the variables of interest, while −1 implied a negative correlation. The summary regression statistic further indicates that there is a positive correlation between the individual models and the ERA5 historical data in 1940–2014.
The summary output of the performance of the CMIP6 models for the historical and the ERA5 observed sea surface temperature data is presented in
Figure 13. A cursory look at
Figure 13 demonstrates the fidelity and consistency of the selected models in reproducing the observed ERA5 data, except for CAMS-CSM.
4.2. Test of Significance
The Pearson product-moment correlation coefficient (r) carried out on the CMIP6 models to validate the performance of the models as shown in
Table 3. The statistical analysis carried under SPSS s indicates that ACCESS, CAMS-CSM, CAN-ESM, CMCC, and MCM UA were statistically significant in the 1-tailed significance test with the r values of 0.00, 0.00, 0.00, 0.00, and 0.00 at 0.00 (
p < 0.05), respectively as shown in
Table 4. This implied that the individual models performed well and were therefore suitable for simulating the statistical patterns of SSTs over the Gulf of Guinea.
The model summary output including the ANOVA and the descriptives shown in the study indicates that the F value of 11.61 was statistically significant at (
p < 0.05). This implies that there is a statistically significant difference between the means of the different model outputs (
Table 4 and
Table 5).
The histogram graph in
Figure 14 shows that the CMIP6 model outputs are normal distributions for future trends (2030–2100). The histogram graph was created as one of the assumptions that should be checked before building a forecasting model to ascertain model normality and linearity. It was important to evaluate the goodness of fit and the statistical significance of the estimated SST model outputs of the constructed regression models; the techniques commonly used to verify the goodness-of-fit values of regression models are hypothesis testing, R-squared, and the analysis of the residuals. In this case, the histogram was skewed towards positive correlations suggesting that the CMIP6 model simulation and the observed results were likely to lead to a skill teleconnection between the SST anomalies and West African rainfall climatology.
The output from the CMIP6 model for all the models showed a smooth trend for the future scenarios (2015–2100). The scatter plot in
Figure 15 shows the normal probability plot between the ERA5 observational data and the CMIP6 future SST simulation. The residual plot further shows that the relationship between the ERA5 observational data and the CMIP6 simulation is linear. Therefore, the probability plots indicate that the error terms are indeed normally distributed. In statistics, the P-P plot compares the observed cumulative Distribution Function (CDF) of the standardized residual to the expected CDF of the normal distribution. The normal probability plot indicates whether the residuals follow a normal distribution, in which case, the points follow a straight line, as is evident in
Figure 14.
The results, however, show that the CMIP6 multi-model ensemble mean values (and CVs) for the six indices have similar magnitudes and patterns of variations as the reanalysis data for the Gulf of Guinea, except the MCM-UA model (USA) (
Table 6).
5. Summary and Conclusions
The study evaluated the performances of bias-corrected GCM-CMIP6 models in stimulating the historical and future projections of SSTs over the Gulf of Guinea involving ACCESS-CM2 (Australia), CAMS-CSM1-0 (China), CanESM5-CanOE (Canada), CMCC-ESM2 (Italy), HadGEM3-GC31-LL (UK), EC-Earth3-CC (Europe), MCM-UA-1-0 (USA), and MPI-ESM1-2-LR (Germany), while ERA5 reanalysis data were used as the observed/historical data for the validation of the models. Given the fact that GCM spatial resolutions are often too coarse, the bias-corrected dataset was developed using Empirical Quantile Mapping (EQM) for the historical (1940–2014) and projected (2015–2100) periods to achieve reliable projections at the regional and local scales. The CMIP6 GCM model projection further revealed that ACCESS-CM2 CAMS-CSM1-0, CanESM5, CMCC-ESM2, and MCM-UA performed better in reproducing the observed SST climatology. The future SST projections by the models indicate that, on average, SSTs will warm up by 1.03 °C between 2030 and 2040, and a projection of 35 °C by 100 against the historical observation of 30 °C was simulated by the models between 1970 and 2014 for the high-warming region and 21° for the Western Sahel. The model evaluation using the coefficient of regression indicated that all the models, namely, ACCESS, CAMS-CSM, CAN-ESM, CMCC, MUM, and MP-ESM, had R values of 0.10, 0.01, 0.06, 0.09, 0.10, and 0.013, respectively. The weakest correlation was for the CAMS-CSM model, at 0.01, while the strongest correlation was for MCM-UA, at 0.13, which further demonstrated the fidelity of the models in reproducing the observed SST climatology of the Guinean coast.
In view of the foregoing, the study concludes that the eight CMIP6 models that were examined both underestimate and overestimate SSTs over the GOG relative to the ERA5 reanalysis dataset, which is evident over the GOG. The cold bias was mainly focused on over the Western Sahel, which was the most obvious during the cold season, while the warm bias dominated in the Guinean coast. The models were generally successful in representing the spatial variabilities of the climatological mean sea surface temperatures. One of the implications of projected, future SST warming is that it will drive changes in the meridional and zonal SST gradients with a deleterious impact on the local climate and beyond and the location of the Intertropical Convergence Zone (ITCZ), which is a function of the underlying SST and the Equatorial Meridional SST gradients. In addition, small spatial differences in SST warming can also trigger changes in the winds and, hence, rainfall strength and distribution in the future. Finally, the study concludes that the projected increase in future SSTs over the GOG will be higher in the far period end than the near-term climate. Overall, the bias-corrected CMIP6 projections can be used for multiple assessments related to climate and hydrological impact studies and for developing mitigation measures under a warming climate.
The authors are grateful to the two anonymous reviewers for their time and effort in reviewing this work. These suggestions have made this work much better.