Results & Discussion

Results & Discussion

Cluster analysis 

A cluster analysis was performed on the regional model means to identify the similarities and disparities in patterns and magnitude of change of Mean Annual Temperature  (MAT) and Mean Annual Precipitation (MAP) as projected by the 23 selected GCMs.

The cluster analysis revealed that there were closely related models and highly variating ones. At a distance of 5, the analysis reveals that the ensemble of 23 models was grouped into five clusters (Figure 3). 

Figure 3: Dendrogram of the cluster analysis of the 23 models to project future climate on Africa. The level plots corresponding to the clusters show projected patterns and magnitude of changes in mean annual temperature (°C) and mean annual precipitation (%) for the period 2041 - 2060 of the SSP 2-4.5. Inset map shows the IPCC regions of Africa. 

The groups of models showed similar spatial patterns of MAT increase and MAP changes (Figure 3). The Sahara, Mediterranean, Arabian peninsula, and West-southern Africa will have the highest magnitude of MAT increase in the SSP 2 - 4.5 scenario. Group 5 models showed the highest magnitude of change in MAT and MAP. Group 3 models showed the least magnitude of MAT increase, especially in sub-Saharan Africa but overall Group 1 showed the least magnitude of change in both MAT and MAP. The models projected that most regions would have small changes in precipitation except for regions of the Sahara Mediterranean and Southern Africa. regions of Western Africa (WAF) and Central Africa (CAF) would experience substantial increase in MAP.  

Principal Component Analysis

Three Principal Components (Figure 4) explained over 80% of the variance in the model projections of anomalies in mean annual temperature and mean annual precipitation anomalies in 2041-2060. The red vectors represent regional MAT and the blue vectors represent regional MAP.

Mean annual temperature (MAT) increased for all regions and high values with majority of variance explained by majorly principal component 1 (Comp.1). The high MAT values were also more associated with models from the clusters of group 5 (CanESM5-CanOE and CanESM5) and group 4 (ACCESS-ESM1-5, ACCESS-CM2, UKESM1-0-LL and HadGEM3-GC31-LL. Mean annual precipitation (MAP) increased for all regions except the Madagascar (MDG) and Arabian Peninsula (ARP) and variance was more explained by principal omponent 2 (Comp.2). MAP was more associated with models of group 1 (yellow points) (IPSL-CM6A-LR, FIO-ESM-2-0, CMCC-ESM2, CNRM-CM6-1-HR, CNRM-CM6-1, CNRM-ESM2-1, GISS-E2-1-G, GISS-E2-1-H) and group 2 of EC-Earth3-Veg and EC-Earth3-Veg-LR. 


Figure 4 (left): Components 1 and 2 represent 69% of the variation and (right) Components 1 and 3 represent 63% of the variation of regional MAT (red vectors) and MAP (blue vectors) as projected by  the GCMs (scores). The grouped GCMS: group 1 - yellow, group 2 - green, group 3 - pink, group 4 - teal and group 5 - orange.  

Regional model selection

The 23 model ensemble had a continental average MAT of 2.1°C ranging from 1.4 to 2.9 °C however some regions were projected to have increase over 3.3°C such as Sahara (SAH) and Mediterranean (MED). The full ensemble projected continental precipitation to increase by 4% ranging from -1% to 8%. However, some regions were projected to have an over 20% precipitation increase such as Western Africa (WAF) and an over 5% decrease in  MED and West-southern Africa (WSAF). The regional model averages illustrated the "worst case", "best case", and "median" climate change projections scenarios. What models constituted the worst case scenario differed by region and by the climate variable of interest. The 23 model average showed that the Sahara (SAH), Arabian peninsula (ARP), MED, West- and East-southern Africa (WSAF and ESAF) were projected to experience the highest increase in MAT. Madagascar (MDG), MED and WSAF were projected to register reduced precipitation in the 2041-2060 period.

Table 3: Mean annual temperature change (°C) (yellow to orange) and mean annual precipitation change (%) (cream to blue) of the ten IPCC regions of Africa as projected by 23 CMIP6 GCMs sorted by the clustered groups. 

The interaction between decreasing or increasing MAP and magnitude of increase in MAT was analyzed for each region to understand the projections for regions with cool and wet conditions as well as those with hot and dry conditions (Table 4).  The cool and wet also the "best" case scenario (green) for Africa (Table 4 last column) was projected by by INM-CM5-0 and INM-CM4-8 whereas the hot and dry also the "worst" case scenario (red) is projected by UKESM1-0-LL and HadGEM3-GC31-LL. It was also noted that the model projections for each region were in the same range implying that the direction of change for the regions was clear. 

Table 4: Projection of the "best" (wet and cool) and "worst" (hot and dry) case scenarios for the ten IPCC regions and Africa as projected by 23 GCMs sorted by best to worst for Africa. 

Discussion

I selected 23 General Circulation Models of CMIP6 to project climate change expressed by mean annual temperature (MAT) and mean annual precipitation (MAP) of the ten IPCC regions of Africa for the future period 2041-2060 under SSP2-4.5. The models had varying projections for climate change and were clustered into five groups. Group 1 models: IPSL-CM6A-LR, FIO-ESM-2-0, CMCC-ESM2, CNRM-CM6-1-HR, CNRM-CM6-1, CNRM-ESM2-1, GISS-E2-1-G, GISS-E2-1-H estimated the least change in MAT and MAP and group 5 models CanESM5-CanOE and CanESM5 estimated the highest change in MAT and MAP. The high climate sensitivity of CanESM5 models UKESM1-0-LL have been noted elsewhere as in  (Mahony, Wang et al. 2022). Nonetheless, I would encourage their consideration for the continent because their projected MAT increase did not surpass the unlikely climate sensitivity threshold of 4.5°C cited by (Sherwood, Webb et al. 2020). The spatial patterns of MAP and MAT in this study are similar to projections by the Climate Assessment Report by IPCC (Gutiérrez, Jones et al. 2021, Iturbide, Fernández et al. 2021). The Sahara (SAH), Arabian peninsula (ARP), Mediterranean (MED), West- and East-southern Africa (WSAF and ESAF) were projected to experience the highest increase in MAT. Mean precipitation is also expected to decrease with medium to high confidence in regions of East-southern Africa (ESAF), Mediterranean (MED) and West-southern Africa (WSAF) (Gutiérrez, Jones et al. 2021, Iturbide, Fernández et al. 2021).

The cluster analysis grouped models from the same institutions together confirming that these usually incorporate the same set of processes at similar resolutions (Knutti et al., 2010). The cluster analyis used simple averages however weighted averages (using Bayesian, orthogonal or other methods) based on past relationships between forecasts and verifications are reported to be more reliable (Knutti, Furrer et al. 2010, Ashfaq, Rastogi et al. 2022).

Important to note is that model projections for the continent and its regions have previously shown some bias especially for precipitation patterns. Group 5 models (Figure 3) showed wetter conditions in the SSP2-4.5 of higher levels of greenhouse gases and yet recent observations demonstrate there are increasing in droughts in East and Southern Africa (Carvalho, Santos et al. 2017, James, Washington et al. 2018).  However, this maybe further analyzed by use of seasonal parameters instead of the mean annual precipitation as was in (Endris, Omondi et al. 2013). Most studies in Africa substantiated that GCMs provided better estimates for temperature than precipitation (Endris, Omondi et al. 2013, Gbobaniyi, Sarr et al. 2014, James, Washington et al. 2018).

Selected regional model ensembles are increasingly being used in preference to single GCMs for example (Mahony, Wang et al. 2022) for North America.   A study of 10 models for mean annual temperature and precipitation over west Africa by (Gbobaniyi, Sarr et al. 2014) also confirms that Multi-model projections show lower biases than single models. Selection of model ensembles was also hinged on selecting the appropriate number of representative models as reviewed by (Raju and Kumar 2020). (Mahony, Wang et al. 2022) and (GLISA 2021) resonated to the selection of at least eight models for regional or extensive studies. The selection of representative and meaningful climate variables were also good practices for climate modelling using GCMs.

For the case of best fitting model or minimum bias, selection of models was done by enacting performance runs using historical observations should be made as was done by (Ahmed, Sachindra et al. 2019) did for 36 CMIP5 models over Pakistan. Regional selections of models has greatly varied from region or country studied in the past for example (Shiru, Chung et al. 2020) in Nigeria selected HadGEM2-ES, CESM1-CAM5, CSIRO-Mk3.6.0, and MRI-CGCM3 while (Assamnew and Tsidu 2020) selected GCMs according to seasonal changes in Eastern Africa.

Conclusion

The use of regional circulation models is increasing and favored to reduce the coarseness of global climate model simulations. Downscaling of these climate products to regional levels improves guidance for actionable climate adaptation and natural resource management initiatives. Similarly, the use of Multi-model ensembles (MMEs) reduces the biases of individual models however, care should be taken in the selection of models for these ensembles.  Like (Whittleston et al. 2017) and (James et al. 2018) I suggest that future studies on climate modeling in Africa consider that important continental climate processes such as the intertropical convergence zone (ITCZ), African easterly jet (AEJ), tropical easterly jet (TEJ), and certain teleconnections are validated to increase confidence in model projections. The findings of this study also agree with previous research that the Mediterranean, Sahara and regions of Southern Africa will experience the hottest and driest climate in the period 2041-2060 in the middle-of-the-road scenario. This study will guide regional model selection for "best" to "worst" case scenarios and also stimulate further research on a comprehensive review of Africa’s regional climate modeling.