Comparison of Indian Ocean warming simulated by CMIP5 and CMIP6 models

LI Jingyi and SU Jingzhi

Institute of Climate System,Chinese Academy of Meteorological Sciences,Beijing,100081,China

ABSTRACT Under the ongoing global warming,the sea surface temperature(SST)over the entire Indian Ocean(IO)has been warming saliently at a rate of 0.014°C yr−1 since the 1950s,which is larger than that in other regions of the globe.The salient IO warming reflects the synergistic effect of global warming and the internal variability of the climate system,and the warming could lead to climate anomalies in peripheral regions.The simulation performance of the sustained IO warming was evaluated by comparing 37 CMIP5 and 37 CMIP6 models with observed data.The results show that the warming in the IO can be captured by nearly all the CMIP models, but most tend to underestimate the magnitude of IO warming trends. There is no qualitative improvement in the simulation of the salient IO warming from CMIP5 to CMIP6. In addition, six metrics were used to investigate the performance of all models.Concerning the spatial pattern of warming trends, the CMIP5 models reveal a better simulation performance than those in CMIP6 models.Only nine best models(seven CMIP5 models and two CMIP6 models) can simulate a high warming trend in the IO region of 0.014 ± 0.001°C yr−1 during 1950-2005, but these nine models still have some disadvantages among other metrics. The overall evaluation here provides necessary information for future investigation about the mechanism of the sustained IO warming based on the climate models with better performances.

KEYWORDS Indian Ocean;salient warming;CMIP5;CMIP6

1. Introduction

The global temperature has warmed significantly for more than 100 years since the Industrial Revolution(IPCC 2013). The temperature warming exhibits great regional differences over the globe, due to the complex coupling processes within the climate system. A unique feature is the salient warming in the Indian Ocean (IO),which has sustained for more than a half century already(Yamagata et al.2004;Ihara,Kushnir,and Cane 2008;Roxy et al.2016).The persistent warming covers the overall IO region since the 1950s with a trend of 0.014°C yr−1,which is much larger than other regions in the global ocean(Figure 1(a)).

The persistent IO warming could have impacts on climate variability over several regions. A warming trend over the IO region would not only directly modify the interannual monsoon variability (Yang et al. 2007;Swapna,Krishnan,and Wallace 2013),but also affect the water vapor transport and precipitation intensity in the monsoon region (Zhang 2001; Choudhury, Nath, and Chen 2019).Besides local responses,the regional warming in the IO also has remote effects.The warm trend in the IO has a potential impact on the positive phase shift of the North Atlantic Oscillation (Hoerling, Hurrell, and Xu 2001)and may influence El Niño events during their developing and terminating phases (Annamalai et al.2005;Kug and Kang 2006).

Many previous studies have investigated the sustained SST warming over the IO (Levitus et al. 2000;Levitus, Antonov, and Boyer 2005; Hoerling et al. 2004;Alory,Wijffels,and Meyers 2007;Yu,Jin,and Weller 2007)and one basic reason is believed to be due to changes in net air-sea heat flux induced by global greenhouse gas concentrations, which are mainly associated with anthropogenic forcing. A few studies have suggested an important role played by ocean advection processes in warming and expansion of the Indian Ocean warm pool (Rao et al. 2012; Dong, Zhou, and Wu 2014).Investigating such half-a-century basin-wide warming relies on both observational data analysis and climate system model results.CMIP has already provided plentiful simulations, which can be used as a baseline to investigate climate variability from multiple views (Du and Xie 2008; Taylor, Stouffer, and Meehl 2012; Eyring et al.2016;Xu,Su,and Zhu 2014;Xu et al.2017).

However, the sustained warming phenomenon over the IO has not been analyzed based on CMIP5 and CMIP6 models yet. Therefore, this study evaluates the simulation performance of the persistent warming over the IO between CMIP5 and CMIP6 models.It is known that the climate sensitivity of some CMIP6 models is stronger than that in CMIP5 (Forster et al. 2020; Zelinka et al.2020), due to updated parameterizations and other model changes. Such an investigation may help us validate the model simulations with observed datasets to better estimate the model uncertainty and provide some suggestions for model improvement.

Figure 1. Spatial distribution of the SST trend in the global region during 1950-2005 (shading; units: °C yr−1):(a) ERSST.v5 dataset(purple contour is the area that passes the 95% confidence test of the warming trend); (b) averaged values of the top five CMIP5 models;(c)averaged values of the top five CMIP6 models;(d)composite of all CMIP6 models.The SNR(2)of the trend is indicated by the green contour for the ratio of the mean value of the trend to the standard deviation of all 37 models. The spatial correlation coefficients and RMSEs between(a)and(b-d)are respectively marked in the lower-left corner of each panel.The IO basin is indicated by the blue box(35°S-25°N,33°-113°E).

2. Data and methods

The ERSST.v5 dataset is used for model validation, which provides monthly data on SST with a resolution of 2.0° ×2.0° (Huang et al. 2017). To compare the simulation performances of CMIP5 and CMIP6 models with respect to the salient warming over the IO, the outputs of the historical simulations by the available 37 CMIP5 models and 37 CMIP6 models are used here. The monthly-mean SSTs from the CMIP5 and CMIP6 models are provided by the Program for Climate Model Diagnosis and Intercomparison(Taylor,Stouffer,and Meehl 2012;Eyring et al.2016).

Since the historical simulations in CMIP5 models are limited to 2005, the period selected in the study is 1950-2005. Comparing the trend distribution of global SST between 1950-2005 and 1950-2014, it shows that similar warming patterns occurred in these two periods(Figure 1(a) and Figure S1). All model products were

compared with the observed trend(Figure 2(a,c)).There are several models in CMIP5 whose warming trends (IO trend metric) are quite close to the observed value(0.014°C yr−1); namely, MPI-ESM-LR, CESM1-WACCM,CMCC-CMS, FGOALS-g2, NorESM1-M, CCSM4, and bcccsm1-1-m. The difference between these seven models and the observed trend is less than 0.001°C yr−1,indicating that the value of the IO warming trend is simulated well by those models.However,the difference between the trend in the IO and other ocean regions (Difftrend metric) indicates a large deviation in the simulations of almost all the CMIP5 models, and only MPI-ESM-LR,FGOALS-g2, and GFDL-ESM2M show a value relatively close to observed data(0.0038°C yr−1).Particularly,there are still some models in CMIP5 where the warming trends are negative, which would reduce the overall simulation performance of CMIP5.

For the IO trend metric in CMIP6 (Figure 2(b)), only two models(EC-Earth3-Veg and CanESM5) have a small deviation(<0.001°C yr−1)from the observed IO warming trend.The result of this metric is not as good as CMIP5.For the metric of the difference between the trend in the IO and in other ocean regions (Difftrend metric) in CMIP6 (Figure 2(d)), six models (BCC-CSM2-MR,NorCPM1, MPI-ESM-1-2-HAM, NESM3, NorESM2-MM,and NorESM2-LM) show values close to the observed value(0.0038°C yr−1).The difference between the trends in the IO and other ocean regions simulated by the models in CMIP6 is relatively better, implying a better performance in this aspect than that in CMIP5 models.

Figure 2.(a)Spatially averaged warming trend in the IO region of the ERSST.v5 dataset and each CMIP5 model.(b)As in(a)but for CMIP6 models.(c)Difference between the spatially averaged warming trend in the IO region and other ocean regions of the ERSST.v5 dataset and the CMIP5 models.(d)As in(c)but for CMIP6 models.All trends have units of°C yr−1.The warming period selected is 1950-2005.

between the IO trend and other ocean regions simulated by this model is the most consistent with the observed value. Among the two metrics, the difference between the simulated value and the observed may be positive or negative, and its absolute value was considered here.For the correlation coefficients metrics (Corr IO metric and Corr glb metric),a zero value(the dark blue square in Figure 4)means that the correlation coefficient of the simulated trend and the observed value in this region(IO/global) is the maximum value among all models in CMIP5 and CMIP6. A larger correlation coefficient indicates a higher correlation between the simulated and observed values,suggesting a better performance of the model simulation. Similarly, if the metrics of RMSEs(RMSE IO metric and RMSE glb metric) are zero (the dark blue square in Figure 4), the RMSEs of the trends between the simulated and the observed values in this region(IO/global)are the minimum value of all RMSEs in CMIP5 and CMIP6. Smaller RMSEs indicate smaller error between the simulated and observed values, and a better performance of the model simulation.

Figure 4. Simulation performance metrics for each model of (a) CMIP5 and (b) CMIP6. Six primary metrics are used to depict the significant warming in the IO region:IO trend(the warming trend in the IO region);Difftrend(the difference between the warming trends of the IO and other ocean regions);Corr IO(the correlation coefficients simulated by each model against the observed SST trend in the IO);Corr glb(the correlation coefficients simulated by each model against the observed SST trend in the global ocean);RMSE IO(the RMSEs simulated by each model against the observed SST trend in the IO);and RMSE glb(the RMSEs simulated by each model against the observed SST trend in the global ocean).The warming score is defined as the average of these six metrics.

Among all 37 models of CMIP5,there are five with the best scores (GISS-E2-H-CC, CESM1-WACCM, GFDLESM2G, FIO-ESM, and GISS-E2-H). The top five models are also selected among the CMIP6 models (MCM-UA-1-0, CIESM, GISS-E2-1-G, FGOALS-f3-L, and MPI-ESM1-2-LR). The scores of the top five CMIP5 models are smaller than those of the top five models in CMIP6,indicating that some models in CMIP5 have comparably higher performance in simulating the sustained warming of the IO than CMIP6 models.If being ranked according to the mean score of five metrics for all 74 models,GISS-E2-H-CC,CESM1-WACCM,and GFDL-ESM2G are the best,followed by FIO-ESM,GISS-E2-H,which are all from CMIP5.In general,however,some other models in CMIP5 also have larger errors than CMIP6.Overall, the analysis of these metrics depicts that the skills of individual models in simulating IO warming in CMIP5 are generally better than in CMIP6. However, more attention should be paid to the fact that the selected warming spatial range or other factors may affect the metric values,and further work is still needed to elucidate this finding.

4. Conclusions and discussion

Since the 1950s,the IO basin has experienced consistent warming that lasted for half a century, which may exert a profound impact on the natural environment and climate change(Han et al.2014;Roxy et al.2016).This study evaluated the salient warming phenomenon in the IO region simulated by 37 CMIP5 models and 37 CMIP6 models.In general,the simulated trends of the IO warming in CMIP6 models show little difference/improvement from those in CMIP5 models. Most of these models can capture the warming in the IO region, but have a tendency to underestimate the magnitude of IO warming trends, especially the CMIP6 models. Some CMIP5 models have a better representation in terms of salient warming in the IO.The differences between the IO trend and other ocean regional trends is slightly greater in CMIP6 than CMIP5 models, which indicates that the salient IO warming relative to other regions can be generally well simulated in the CMIP6 models. However, both the RMSE and correlation coefficient metrics show a better simulation performance by CMIP5 models. Regardless,there is no qualitative improvement in the simulations of the IO warming from CMIP5 to CMIP6. The cause of the better simulation of CMIP5 models has not been clarified yet. Analysis of a few studies suggests that the differences between CMIP5 and CMIP6 models may be attributable to generally greater climate sensitivity in the CMIP6 model ensemble relative to the CMIP5 ensemble(Forster et al.2020;Zelinka et al.2020),in part because of developments in the representation of cloud physics and the updated parameterizations. Another factor may be related with the multidecadal oscillation in the climate system, whose observed time phases can barely be well captured by current models.

It should be emphasized that the persistent IO warming is one unique phenomenon under global warming.Driven by global warming, the warming trend in any region would be irreversible sooner or later. However,the warming trend in the IO was quite salient compared with other regions during the past half-century.Besides analyses using observed datasets, investigations based on simulations/experiments with climate models could provide valuable clues for the reasons behind the persistent and significant warming in the IO. It is desirable that further research is carried out with a focus on the underlying mechanism of the sustained IO warming.To achieve such a goal,more analyses and experiments may be needed based on the climate models with a better performance for the sustained IO warming, as given based on the overall evaluation in this study.

Acknowledgments

The authors wish to thank the two anonymous reviewers for their constructive and helpful comments. The ERSST.v5 data were obtained from http://www.esrl.noaa.gov. The CMIP6 model data and CMIP5 model data are available from the ESGF at https://esgf-node.llnl.gov/search/.

Disclosure statement

No potential conflict of interest was reported by the authors.

Funding

This work was supported by the National Key R&D Program of China[grant numbers 2016YFA0600602 and 2019YFC1510002]and the National Natural Science Foundation of China [grant number 41776039].