The present study assessed the performance of GCMs of CMIP5 and CMIP6 in simulating historical annual rainfall, Tmax and Tmin over Egypt for the period 1979−2005, and their projections for the futures. An arid and hyper-arid climate dominates Egypt. Scarce rainfall and high temperature have made the country one of the most water-stressed regions in the world. Moreover, the friable environment has made the country one of the climate change hotspots of the world. Performance of the CMIP5 and CMIP6 GCMs are evaluated in this study to aid climate service in this global climate hotspot.
Improvement in CMIP6 models' performance compared to CMIP5 models has been reported across the globe (Gusain et al. 2020; Zhu and Yang 2020; Kamruzzaman et al. 2021; Song et al. 2021c; Su et al. 2021). Zhu and Yang (2020) reported a higher ability of CMIP6 MME to reconstructing historical rainfall in the dry region of the Tibetan Plateau. Su et al., (2021) showed less bias in estimated droughts using CMIP6 GCMs in China. Ga et al., (2021) reported better capacity of CMIP6 GCMs in reproducing southern hemisphere westerlies. However, the lower or similar performance of CMIP6 GCMs compared to CMIP5 GCMs has also been reported in several studies (Srivastava et al. 2020; Zhu and Yang 2020; Song et al. 2021a). Zhu and Yang (2020) showed lower performance of CMIP6 MME in the wet region of the Tibetan Plateau. Song et al., (2021a) showed higher uncertainty in projections for SSP scenarios compared to RCP scenarios. Srivastava et al., (2020) showed similar bias in both CMIPs in reproducing climatology over the United States. The improvement or reduction in performance was also different for different regions. Therefore, it is important to assess the relative performance of the CMIP5 and CMIP6 GCMs at a regional scale. Besides, it is important to update the climate projections based on new scenarios to realigning the adaptation measures with new projections.
The present study evaluated the performance of 13 common CMIP5 and CMIP6 models. The performance was evaluated for both the individual models and their MME mean. In addition, the performance of individual models in replicating historical climatology was evaluated using KGE, considering its capability to estimate three statistical indices (correlation, bias and variability) together. The results revealed improvement of most CMIP6 GCMs compared to their counterpart in CMIP5 in simulating rainfall and temperature in Egypt.
The MME of CMIP5 and CMIP6 GCMs were compared to evaluate their ability to estimate spatial and seasonal variability of rainfall and temperature. The CMIP6 GCM MME was better capable of replicating the seasonal variability of both rainfall and temperature. The uncertainty in seasonal variability of climate was much less for CMIP6 compared to CMIP5. This was particularly true for rainfall and Tmin. Both MMEs showed wet bias in dry season rainfall. However, the bias was less for CMIP6 MME. Winter rainfall, which shares the major portion of total annual rainfall in Egypt, was reliably captured by CMIP6 MME, while CMIP5 MME showed a slight underestimation. Both the MME underestimated Tmax, but the underestimation was less for CMIP6 MME. The seasonal variability Tmin was also reliably captured by CMIP6 MME. Both the CMIP5 and CMIP6 MMEs could reproduce the spatial pattern of rainfall climatology over Egypt reasonably. However, both the MMEs underestimated rainfall in the northern high rainfall region and overestimated rainfall in the southern low rainfall region. But the biases are less for CMIP6 MME compared to CMIP5 MME.
The results collaborate with the findings in the nearby region. Bağçaci et al. (2021) evaluated the performance of CMIP6 and CMIP5 GCMs in Turkey. They used the MME of only the top four GCMs of both the CMIPs for evaluation. The results showed 11% less bias in rainfall and 6% less bias in temperature in CMIP6 MME compared to CMIP5 MME. Ayugi et al. (2021) compared CMIP6 and CMIP5 models in simulating mean and extreme rainfall over East Africa and reported less bias in CMIP6 MME. Zhu and Yang (2021) evaluated interannual characteristics of CMIP5 and CMIP6 rainfall simulation over North Africa and also showed better performance of CMIP6 models. Studies in other regions also showed better performance of CMIP6 GCMs compared to CMIP5 GCMs.
The present study showed that not all the CMIP6 GCMs have improvements over the CMIP5. Some of the CMIP6 GCMs showed poorer performance in replicating rainfall properties compared to their CMIP5 counterpart. MIROC5 compared to MIROC6 in simulating historical rainfall. CanESM of CMIP5 also showed better performance compared to its counterpart of CMIP6. However, it was noticed that no model was the best performing model for all the three climate variables. For example, CanESM was found the best for simulating Tmax and rainfall but was the worst for Tmin. Overall, the performances of CMIP6 GCMs and their MME were better than CMIP5 in Egypt.
Improvement in model resolution from one generation of CMIP to another generation improved the model's performance. For example, the spatial resolution of many CMIP5 models was higher than CMIP3 models. Sun et al., (2015) evaluated the performance of CMIP3 and CMIP5 GCMs. They concluded that improvement in CMIP5 models' skill over CMIP5 models was partially due to spatial resolution improvement. Improved parameterizations and additional process representations improve models' spatial resolution and eventually improve models' skills (Sheffield et al. 2013). The CMIP6 GCMs' resolutions are not different from CMIP5. Therefore, the performance of CMIP6 GCMs is not much different from CMIP5 GCMs in most of the globe (Rivera and Arnould 2020; Chen et al. 2021; Yazdandoost et al. 2021). The improved performance of some of the CMIP6 GCMs may be due to enhanced parameterization. The present study revealed significant improvement in CMIP6 GCMs and their MME in replicating spatial distribution and seasonal variability of climate in Egypt than CMIP5.
In this study, GCMs were gridded into a resolution of 1º×1º for intercomparison. In literature, the performance of GCMs is evaluated for different grid spacings. CMIP5 GCMs have been re-gridded to 2° ×2° resolution for comparison in a large number of studies as it was nearly equal to the mean grid size of CMIP5 GCMs (Salman et al. 2018; Ahmed et al. 2019, 2020; Iqbal et al. 2020; Khan et al. 2020). It has also been re-gridded to the most common resolution of investigated GCMs (Raju and D. Kumar 2014; Raju et al. 2017). However, CMIP5 and CMIP6 GCMs were mostly re-gridded to the resolution of 1º×1º. For example, they were re-gridded to finer resolution (1°×1°) for comparison of their skill in South America (Rivera and Arnould 2020), East Asia (Chen et al. 2021), Iran (Yazdandoost et al. 2021) and south pacific oscillation (Wang et al. 2021). Varieties of statistical metrics have been used for GCM performance evaluation. The GCM evaluation of performance based on a single metric is always disputed as each metric reflects one aspect of the model performance. Therefore, the present study employed KGE, which can estimate multiple properties, including association, bias and similarity in variability together.
The CMIP6 MME showed a higher rise in temperature compared to CMIP5 MME over Egypt. Both the Tmax and Tmin will rise 0.5°C more compared to that found using CMIP5 model in earlier studies (Nashwan and Shahid 2020). In addition, the decrease in rainfall in the northern high rainfall region, where most of the agriculture activities concentrated, was projected more by CMIP6 MME compared to CMIP5 MME. This indicates a much worse situation in the future in the country compared to that anticipated earlier using CMIP5 models (Nashwan and Shahid 2020). Similar results have been reported by Almazroui et al. (2020) over Africa using CMIP6 models. They showed median warming of CMIP6 ensemble higher than CMIP5.