Surface Air Temperature Simulations over China with CMIP5 and CMIP3

Abstract

Historical simulations of annual mean surface air temperature over China with 25 CMIP5 models were assessed. The observational data from CRUT3v and CN05 were used and further compared with historical simulations of CMIP3. The results show that CMIP5 models were able to simulate the observed warming over China from 1906 to 2005 (0.84°C per 100 years) with a warming rate of 0.77°C per 100 years based on the multi-model ensemble (MME). The simulations of surface air temperature in the late 20th century were much better than those in the early 20th century, when only two models could reproduce the extreme warming in the 1940s. The simulations for the spatial distribution of the 20-year-mean (1986–2005) surface air temperature over China fit relatively well with the observations. However, underestimations in surface air temperature climatology were still found almost all over China, and the largest cold bias and simulation uncertainty were found in western China. On sub-regional scale, northern China experienced stronger warming than southern China during 1961–1999, for which the CMIP5 MME provided better simulations. With CMIP5 the difference of warming trends in northern and southern China was underestimated. In general, the CMIP5 simulations are obviously improved in comparison with the CMIP3 simulations in terms of the variation in regional mean surface air temperature, the spatial distribution of surface air temperature climatology and the linear trends in surface air temperature all over China.

Keywords

CMIP5 ; CMIP3 ; China ; annual mean surface air temperature ; historical simulation ; assessment

1. Introduction

Coupled global circulation models (CGCMs) are useful tools in the simulation and projection of climate change. However, its reliability suffers from a series of uncertainties stemmed from the initial conditions, external forcing agents, etc. Under the World Climate Research Programme (WCRP) the Working Group on Coupled Modeling (WGCM) established the Coupled Model Intercomparison Project (CMIP) as a standard experimental protocol for studying the outputs of CGCMs. The WCRP’s WGCM agreed to promote a new set of coordinated climate model experiments [ Taylor et al. , 2012 ], comprising the fifth phase of the CMIP (CMIP5). Historical simulations are fundamental experiments of great significance, because their comparisons with available observations are a prerequisite to any application of climate model outputs in climate change studies, especially the projection of future climate. Recently, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) has collected and released the historical simulations of more than 40 CMIP5 models, supporting the evaluation of state-of-the-art climate models.

The global mean surface air temperature has experienced a significant increase during the past century. The changes in surface air temperature on small spatial scales, like sub-continent or regional scale, show different features. In comparison with the global mean climate, regional climate has more complex variability since it is influenced by ocean-atmospheric circulation, land cover, and associated feedback processes. Regional climate is relevant to the environment and economic production, and deserves more attention. It is reported that coupled climate models are able to provide relatively reliable simulations for large-scale climate changes, while simulations of regional climate involve large uncertainties [ Zhao and Luo , 1998 ]. Using CMIP3 models, assessment of historical simulations of surface air temperature over East Asia and China have been widely performed [ Zhou and Yu, 2006 ; Xu et al ., 2007  ;  Liu and Jiang, 2009 ]. In comparison with the 20th century historical run of CMIP3, the CMIP5 is forced by more realistic external forcing agents including a time-varying forcing with changes on an inter-annual scale. In addition to the external forcing, most of the CMIP5 models have been significantly improved based on their earlier versions by enhancing their resolutions. Updated physical processes are integrated to describe the prognostic aerosols, atmospheric chemistry, dynamic vegetation and carbon cycle [ Taylor et al ., 2012  ;  Voldoire et al ., 2012 ]. There is a need for the evaluation whether the CMIP5 historical simulations of changes in surface air temperature over China have improved. Although Xu and Xu [2012] have evaluated the CMIP5 simulations of temperature and rainfall in China for the last 45 years (1961–2005), there is no comparation of the CMIP5 simulations with the earlier CMIP3 simulations. In this paper, we intend to evaluate the CMIP5 simulations of surface air temperature in China for the past 100 years (1906–2005) by using the output of 25 CMIP5 models, and then compare the simulations with the earlier generation of coupled climate models.

2. Models and data

In CMIP5, the historical run is performed from 1850 to 2005 with time-varying forcing agents consistent with the observations, including anthropogenic greenhouse gases, short-lived natural and anthropogenic aerosols or their precursors, ozone, volcanic effects, solar irradiance, and land use. Twenty-five models participating in the CMIP5 were assessed in this study. Although CMIP5 models have a set of ensemble members with different initialization methods, initial conditions, and physical parameterization schemes, only the first ensemble member (r1i1p1) was assessed in this paper. The first run (run1) of the 20th century simulation based on 20 CMIP3 models covering the period of 1901–1999 was also evaluated for comparison purposes. The outputs of CMIP5 and CMIP3 models were derived from the PCMDI (http://www-pcmdi.llnl.gov/ ), and their primary information are summarized in Table 1 .

Table 1. Information of CMIP5/CMIP3 climate models used in this study

CMIP5			CMIP3
Model	Country	Resolution	Model	Country	Resolution
BCC-CSM1-1	China	2.8°×2.8°	BCCR-BCM2.0	Norway	2.8°×2.8°
BNU-ESM	China	2.8°×2.8°	CCSM3	USA	1.4°×1.4°
CanESM2	Canada	2.8°×2.8°	CGCM3.1-T47	Canada	3.75°×3.75°
CCSM4	USA	1.25°×0.94°	CGCM3.1-T63	Canada	2.8°×2.8°
CESM1-WACCM	USA	2.5°×1.875°	CNRM-CM3	France	2.8°×2.8°
CMCC-CM	France	0.75°×0.75°	CSIRO-Mk3.5	Australia	1.875°×1.875°
CNRM-CM5	France	1.4°×1.4°	ECHAM5	Germany	1.875°×1.875°
CSIRO-Mk3–6-0	Australia	1.875°×1.875°	FGOALS-g1.0	China	2.8°×3.0°
FGOALS-s2	China	2.8°×1.7°	GFDL-CM2.0	USA	2.5°×2.0°
FIO-ESM	China	2.8°×2.8°	GFDL-CM2.1	USA	2.5°×2.0°
GFDL-ESM2G	USA	2.5°×2.0°	GISS-AOM	USA	4.0°×3.0°
GISS-E2-H	USA	2.5°×2.0°	GISS-EH	USA	5.0°×4.0°
GISS-E2-R	USA	2.5°×2.0°	GISS-ER	USA	5.0°×4.0°
HadCM3	UK	3.75°×2.5°	INM-CM3.0	Russia	5.0°×4.0°
INMCM4	Russia	2.0°×1.5°	IPSL-CM4	France	3.75°×2.5°
IPSL-CM5A-LR	France	3.75°×1.875°	MIROC3.2-hires	Japan	1.125°×1.125°
IPSL-CM5A-MR	France	2.5°×1.25°	MIROC3.2-medres	Japan	2.8°×2.8°
IPSL-CM5B-LR	France	3.75°×1.875°	MRI-CGCM2.3.2	Japan	2.8°×2.8°
MIROC-ESM	Japan	2.8°×2.8°	UKMO-HadCM3	UK	3.75°×2.5°
MIROC-ESM-CHEM	Japan	2.8°×2.8°	UKMO-HadGEM1	UK	1.875°×1.25°
MPI-ESM-LR	Germany	1.9°×1.9°
MPI-ESM-MR	Germany	1.875°×1.875°
MPI-ESM-P	Germany	1.9°×1.9°
MRI-CGCM3	Japan	1.1°x1.1°
NorESM1-M	Norway	2.5°×1.875°

The observed gridded surface air temperature data (0.5°×0.5° grid) was derived from the Climate Research Unit (CRU) high-resolution monthly surface air temperature dataset (CRUT3v) for the period of 1901–2009 (www.cru.uea.ac.uk/cru/data/hrg/ ). The reliability of the CRU dataset to describe the regional scale feature in China is argued because the observational stations in China used for generating the gridded dataset is insufficient. As a supplement, the 1°×1° version of the high-resolution gridded daily dataset CN05 (1961–2005), which was generated by Xu et al. [2009] based on the 751 gauge stations in China, was employed. All observational data and model outputs were interpolated into a common 1°×1° grid using the bi-linear interpolation method. The multi-model ensemble mean (MME) used in this paper was calculated with equal weight.

3. Results

In this study, the historical evolution of annual mean surface air temperature over China simulated by CMIP5 and CMIP3 models was evaluated in terms of three metrics, 1) the evolution of regional-mean surface air temperature, 2) the spatial distribution of surface air temperature climatology, and 3) the spatial distribution of the linear trend of surface air temperature.

3.1. Evolution of regional mean surface air temperature

Figure 1 shows the regional mean surface air temperature anomalies of observation and CMIP5 simulations over China from 1906 to 2005. The corresponding linear trends are summarized in Table 2 . The CRUT3v data reveal that surface air temperature over China has increased during the past 100 years at the rate of 0.84°C per 100 years, and an especially prominent warming is observed in the late 20th century. On inter-decadal time scale, two distinct extremely warm periods are found in the past century: one is the 1940s, and the other is from the 1970s to the present. This inter-decadal variation has been demonstrated by a variety of reconstructed data [ Wang et al ., 1998  ;  Wang and Gong, 2000 ], and is consistent with the observed averages of the Northern Hemisphere and the globe [ Zhou and Yu , 2006 ]. All of the 25 CMIP5 models are able to reproduce the warming phenomenon in China with an average rate of 0.77°C per 100 years based on the MME. However, a large spread is found among the 25 models, especially in the first half of the 20th century. In contrast, the models show enhanced consistency after the 1970s, when all simulations get closer to the observations.

Figure 1. The variation of regional mean surface air temperature anomaly over China from 1906 to 2005 (relative to 1986–2005) of observation (CRUT3v) and CMIP5 simulations (the panel on the right-top are two simulated and one observed curve)

Table 2. The linear trend of regional mean surface air temperature over China from CRUT3v and CMIP3/CMIP5 model simulations (unit: °C per 100 years)

Observation/Model	1906–2005	1901–1999
CRUT3v	0.84	0.64
CMIP3		0.72±0.37
CMIP5	0.77±0.45	0.65±0.42

Zhou and Yu [2006] reported that the CMIP3 models were poor in simulating the 1940s’ extreme warming in China, which was attributed to the shortage of time-varying natural external forcings. Our results show that 92% of the CMIP5 models are still not able to reproduce the extreme warming in the 1940s, although the external natural forcings in the CMIP5 historical runs have considered the variation on interannual scale. Only two simulations, MIROC-ESM and NorESM1-M, can reproduce the observed warming in the 1940s satisfyingly (Fig. 1 ). Based on a set of six integrations of a coupled climate model, Delworth and Knutson [2000] concluded that the early 20th century warming have resulted from a combination of human-induced radiative forcing and an unusually large realization of internal multi-decadal variability of the coupled ocean-atmospheric system. In contrary, Nozawa et al. [2005] deemed that it resulted from the natural external forcings. Our results as derived from the historical runs of the 25 CMIP5 models support the argument of Delworth and Knutson [2000] .

Since the historical runs of some CMIP3 models are ending in 1999, we set the common period as 1901–1999, to compare the two generations of climate models. Table 2 lists the linear trends of observed and simulated surface air temperature over the past century. Compared to the linear trend of the simulated surface air temperature of the CMIP3 MME, the CMIP5 MME shows a similar trend (0.65°C per 100 years) to the observations. However, the CMIP5 simulations have a larger model spread than the CMIP3 simulations.

3.2. Spatial distribution of surface air temperature climatology

The spatial distribution of the 20-year-mean surface air temperature (1986–2005) over China is evaluated among 25 CMIP5 models. The CMIP5 MME can simulate the observed spatial distribution of surface air temperature climatology well (Fig. 2 a). However it underestimates the surface air temperature in most parts of China. Large cold biases are found in Sichuan province, southern Xinjiang and the Tibetan Plateau (Fig. 2 b). To estimate the uncertainty of CMIP5 simulations, we calculate the standard deviation among the 25 CMIP5 simulations for the surface air temperature climatology (Fig. 2 c). It shows that the largest model spread is found in western China, followed by northeastern China, while in southeastern China all models show the highest consistency.

Figure 2. Spatial distribution of the 20-year-mean (1986–2005) surface air temperature from (a) observation (CN05), (b) CMIP5 multi-model ensemble mean minus observation, and (c) the estimated standard deviation of the model spread among 25 CMIP5 simulations (unit: °C)

To quantitatively evaluate the simulation of the spatial distribution of surface air temperature climatology over China, a Taylor diagram [ Taylor , 2001 ] is employed. This is done to visualize the pattern correlation coefficient, the ratio of the simulated pattern standard deviation to the observation, and the pattern root mean square error (RMSE). In Figure 3 , a Taylor diagram is displayed for the simulated spatial distribution of the 20-year-mean temperature (1980–1999) from the CMIP5 and CMIP3 models. It seems that the CMIP5 simulations are generally better than the CMIP3 simulations, as indicated by higher pattern correlation coefficients (distributed around 0.95 in CMIP5 and 0.93 in CMIP3) and smaller RMSE. Among the 25 CMIP5 models, CMCC-CM performs best in simulating the spatial distribution of surface air temperature climatology, followed by MRI-CGCM3, IPSL-CM5A-MR and CCSM4 (Fig. 3 ), which is likely to be attributed to their high resolutions.

Figure 3. Taylor diagrams of the simulated spatial distribution of the 20-year-mean (1980–1999) surface air temperature over China from (a) CMIP5 and (b) CMIP3 models (red dot: multi-model ensemble mean; REF: observation; azimuthal position: spatial correlation coefficient; radial distance: ratio of standard deviation; distance from REF point: root mean square error)

3.3. Spatial distribution of linear trend in surface air temperature

To assess how well the spatial structure of changes in surface air temperature are reproduced with the climate models, the spatial distribution of the linear trend in surface air temperature during 1961–1999 from the CN05 data, CMIP5 MME, and CMIP3 MME are shown in Figure 4 . Surface air temperature has increased all over China during the last 40 years of the 20th century, with a higher warming rate in northern China than in southern China (Fig. 4 a). Although the CMIP5 MME can better reproduce the observed spatial distribution of the linear trend (Fig. 4 b) than the CMIP3 MME (Fig. 4 c), it underestimates the difference of the linear trends between the north and the south.

Figure 4. Spatial distribution of linear trends in surface air temperature for the period 1961–1999 based on (a) CN05, (b) CMIP5 multi-model ensemble mean, and (c) CMIP3 multi-model ensemble mean (unit: °C per 100 years)

The quantitative comparison of the spatial distribution of linear trends in surface air temperature simulated with CMIP5 and CMIP3 models for the period of 1961–1999, using a Taylor diagram, is displayed in Figure 5 . In general, the CMIP5 simulations are superior to the CMIP3, although a large model spread exists. About 28% of the CMIP5 models, against 10% of the CMIP3 models, have exceeded a pattern correlation coefficient of 0.5. An underestimation of the spatial difference of the linear trend is found in both CMIP5 and CMIP3 simulations, with CMIP3 showing smaller differences than CMIP5.

Figure 5. Same as Figure 3 , but for the simulated spatial distribution of the linear trends in surface air temperature during 1961–1999

4. Conclusions and discussion

In this study, the historical simulations of annual mean surface air temperature from 25 CMIP5 models over China were evaluated with the observational data of CRUT3v and CN05. A further comparison was performed with simulations of 20 CMIP3 models. The major findings are summarized as follows.

(1) The CRUT3v data shows that the mean surface air temperature over China has increased at the rate of 0.84°C per 100 years from 1906 to 2005. All of the 25 CMIP5 models were able to simulate the warming in China, and their MME provided a warming rate of 0.77°C per 100 years. The simulations of surface air temperature in the early 20th century were not as good as in the late 20th century. Only two models could reproduce the extreme warming in the 1940s. In comparison with the CMIP3 MME, the CMIP5 MME was more reliable for the warming over the past 100 years, but with a larger spread among the models.

(2) The CMIP5 models can reproduce the observed spatial distribution of surface air temperature climatology (1986–2005 means) to a certain degree. The CMIP5 models performed generally better than the CMIP3 models. However, the CMIP5 models underestimated the surface air temperature climatology in most parts of China. The largest cold bias was found in southwestern China, where the simulation uncertainty was also the largest.

(3) On sub-regional scale, northern China has experienced stronger warming during 1961–1999 than southern China. A large spread was found among CMIP5 models in simulating the spatial distribution of the linear trends in surface air temperature. In general, the CMIP5 models showed higher skills in simulating the spatial distribution of the linear trends in surface air temperature than the CMIP3 models, but were still insufficient in capturing the actual spatial difference of the linear trends.

In general, the CMIP5 simulations are improved in comparison with the CMIP3 simulations, although with larger spread among models. This is likely attributed to the complication of the new generation of climate models and the external forcing agents, such as integrating atmospheric chemistry to describe more detailed process of aerosols, which increases uncertainty. In this study, only simple comparisons are presented, no in-depth comparison are involved, like internal physical processes, external forcing agents, etc. Due to the uncertainty in historical simulations with CMIP5 models, it is very difficult to reliably project the future climate change over China. How to investigate and apply more efficient methods of multi-model mean to extract useful information, such as the upgraded reliability ensemble mean method by Xu et al. [2010], is of great significance in the projection of future climate change over China.

Acknowledgements

We acknowledged the international modeling groups for providing their data, the Program for Climate Model Diagnosis and Inter-comparison (PCMDI) for collecting and achieving the model data, the World Climate Research Programme’s (WCRP’s) Coupled Model Inter-comparison Project for organizing the model data analysis activity. This work was supported by the 973 Program (No. 2010CB950501).

References

1. Delworth and Knutson, 2000 T.L. Delworth, T.R. Knutson; Simulation of early 20th century global warming; Science, 287 (2000), pp. 2246–2250
2. Liu and Jiang, 2009 M. Liu, Z.-H. Jiang; Simulation ability evaluation of surface temperature and precipitation by thirteen IPCC AR4 coupled climate models in China during 1961–2000; Journal of Nanjing Institute of Meotorology (in Chinese), 32 (2009), pp. 256–268
3. Nozawa et al., 2005 T. Nozawa, T. Nagashima, H. Shiogama, et al.; Detecting natural influence on surface air temperature change in the early twentieth century; Geophys. Res. Lett., 32 (2005) L20719
4. Taylor, 2001 K.E. Taylor; Summarizing multiple aspects of model performance in a single diagram; J. Geophys. Res., 106 (2001), pp. 7183–7192
5. Taylor et al., 2012 K.E. Taylor, R.J. Stouffer, G.A. Meehl; An overview of CMIP5 and the experiment design; Bull. Amer. Meteor. Soc, 93 (2012), pp. 485–498
6. Voldoire et al., 2012 A. Voldoire, E. Sanchez-Gomez, D. Salasy Mélia, et al.; The CNRM-CM5. 1 global climate model: Description and basic evaluation; Clim. Dyn. (2012), pp. 1–31
7. Wang and Gong, 2000 S.-W. Wang, D.-Y. Gong; Enhancement of the warming trend in China; Geophys. Res. Lett., 27 (2000), pp. 2581–2584
8. Wang et al., 1998 S.-W. Wang, J.-L. Ye, D.-Y. Gong, et al.; Construction of mean annual temperature series for the last one hundred years in China; Quarterly Journal of Applied Meteorology (in Chinese), 9 (1998), p. 392401
9. Xu et al., 2007 C.-H. Xu, X.-Y. Shen, Y. Xu; An analysis of climate change in East Asia by using the IPCC AR4 simulations; Advances in Climate Change Research (in Chinese), 3 (2007), pp. 287–292
10. Xu and Xu, 2012 Y. Xu, C.-H. Xu; Preliminary assessment of simulations of climate changes over China by CMIP5 multi-models; Atmos. Oceanic Sci. Lett., 5 (2012), pp. 489–494
11. Xu et al., 2010 Y. Xu, X.-J. Gao, F. Giorgi; Upgrades to the reliability ensemble averaging method for producing probabilistic climate-change projections; Climate Research, 41 (2010), pp. 61–81
12. Xu et al., 2009 Y. Xu, X.-J. Gao, Y. Shen, et al.; A daily temperature dataset over China and its application in validating a RCM simulation; Adv. Atmos. Sci., 26 (2009), pp. 763–772
13. Zhao and Luo, 1998 Z.-C. Zhao, Y. Luo; Advance on investigations of regional climate modelling since 1990; Acta Meteorologica Sinica (in Chinese), 56 (1998), pp. 225–241
14. Zhou and Yu, 2006 T.-J. Zhou, R.-C. Yu; Twentieth-century surface air temperature over China and the globe simulated by coupled climate models; J. Climate, 19 (2006), pp. 5843–5858