LIN Peng-Fei, FENG Xiao-Li, and LIU Juan-Juan

1State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics (LASG), Institute of Atmospheric Physics (IAP), Chinese Academy of Sciences, Beijing 100029, China

2Meteorological Observatory, Huangnan Tibetan Autonomous Prefecture of Qinghai Province, Huangnan Tibetan Autonomous Prefecture 811300, China

3Ministry of Education Key Laboratory of Meteorological Disaster of Cooperation of Ministries and Provincial Governments and College of Atmospheric Sciences, Nanjing University of Information Science and Technology, Nanjing 210044, China

Historical Trends in Surface Air Temperature Estimated by Ensemble Empirical Mode Decomposition and Least Squares Linear Fitting

LIN Peng-Fei1, FENG Xiao-Li2,3, and LIU Juan-Juan1

1State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics (LASG), Institute of Atmospheric Physics (IAP), Chinese Academy of Sciences, Beijing 100029, China

2Meteorological Observatory, Huangnan Tibetan Autonomous Prefecture of Qinghai Province, Huangnan Tibetan Autonomous Prefecture 811300, China

3Ministry of Education Key Laboratory of Meteorological Disaster of Cooperation of Ministries and Provincial Governments and College of Atmospheric Sciences, Nanjing University of Information Science and Technology, Nanjing 210044, China

Ensemble empirical mode decomposition (EEMD) and least squares linear fitting (LSLF) are applied to estimate the historical trends of surface air temperature (SAT) from observations and Coupled Model Intercomparison Project Phase 5 (CMIP5) simulations during the period 1901-2005. The magnitudes of trends estimated by the two approaches are comparable. The trend calculated by the EEMD approach is larger than that by the LSLF approach in most (23/27) of the models during 1901-2005. During the slow warming period, the EEMD trend is smaller than the LSLF trend. The rootmean-square errors (RMSEs) between the raw and reconstructed times series by the LSLF approach are larger than those by the EEMD trend component and multi-decadal variability components during 1901-2005 in most of the models and observations. During 1901-70 (or 1971-2005), the RMSEs between the raw and reconstructed times series by LSLF are larger than those by the EEMD trend component. In this sense, the EEMD trend is a better choice to obtain the climate trends in observations and CMIP5 models, especially for short time periods. This is because the trend estimated by LSLF cannot capture the internal variability and the cooling in some years. The estimated global warming rates (trend) are consistently larger (smaller) than those from observations in 11 of 27 CMIP5 models during 1901-2005 in the slow and rapid warming periods. This implies these 11 models have consistent responses to greenhouse gases for any period.

climate trend, surface air temperature, CMIP5, LSLF, EEMD

1　Introduction

Determining the climate trend is an important step in statistics and numerous scientific analyses, and is key for estimations of global warming. Most studies addressing possible climate trends in climate datasets assume a priori that the trends are linear and consequently use leastsquares linear fitting (LSLF) to extract the trend (e.g., Born, 1996; Dai et al., 1997; Norris, 2005; Stocker et al., 2013). Linear trend analysis can estimate the quantitative trend in a simple way. However, estimating the climate trend using the linear approach is not physically realistic and can lead to uncertainty because the temporal behavior is complex in a changing climate (e.g., Wu et al., 2011; Qin et al., 2012). The ensemble empirical mode decomposition (EEMD) approach (Wu and Huang, 2009) can extract the temporally changing nonlinear trends (Franzke, 2009, 2010, 2012; Wu et al., 2011) from a changing complex climatic time series.

Coupled models are often used to detect/attribute historical observed trends by testing alternative scenarios driven by different forcings (e.g., natural, anthropogenic) (Jones et al., 2013; Knutson et al., 2013). The first step is to access and estimate the historical trends precisely from coupled climate models. However, due to biases and internal variability from different models, using just one model can bring relative large uncertainty. To combat this problem, multiple models are instead often used. The Coupled Model Intercomparison Project Phase 5 (CMIP5) has provided simulation results from multiple models. Evaluating the CMIP5 simulations using EEMD can help us understand the simulation abilities.

The aims of this study are to estimate and assess the historical climate trends from multiple CMIP5 models and observations using the EEMD and LSLF methods, and compare the surface air temperature (SAT) trends in different periods. The temporally changing nonlinear and linear trends are presented during the slow warming period and the rapid warming period. The possible differences in the trend lines and their causes in the slow and rapid warming periods according to the two approaches are explored. A brief description of the data and methods is given in section 2, follows by the results in section 3, and a brief summary in section 4.

2　Data and methods

We select the first ensemble member (i.e., “r1i1p1”) in the historical runs forced by greenhouse gases, sulfate aerosols, and volcanic and solar activity during the period of 1850 (or 1860) to 2005 from 27 CMIP5 models (Tay-lor et al., 2012). Table 1 shows the details of the CMIP5 models, including the assigned model number for each model name, the modeling center, the affiliated country, and the reference (or link). SATs simulated by the models are compared with observed SATs from the the National Aeronautics and Space Administration Goddard Institute for Space Studies Surface Temperature Analysis (GISTEMP) (Hansen et al., 1999) and the Met Office Hadley Centre and Climatic Research Unit latest version (HadCRUT4) (Morice et al., 2012) datasets.

The climatological mean values of 1971-2000 are subtracted from the annual mean SATs during 1901-2005. LSLF is used to compute the linear trends of global averaged annual mean SAT anomalies (relative to the mean value from 1971 to 2000). We also compute the linear trends of the global averaged SATs (area-weighted) during the different periods (1901-2005, 1901-70, and 1971-2005) in observations and the historical runs from CMIP5. For calculating the multi-model ensemble (MME) mean, the equal-weight for every model is used. Since the climate trend cannot be linear, we choose the EEMD approach to extract the climate trend. The EEMD approachis also applied to time series of global averaged annual mean SAT anomalies. EEMD is a noise-assisted data analysis approach and it can extract the nonlinear trend if the time series have such a trend. The white noise is added in this approach. The white noise has an amplitude of 0.1 the standard deviation of the original time series. For an each EEMD ensemble member, the ensemble size is 500 times. In this study, the original time series during 1901-2005 is split into six intrinsic mode functions (IMFs, also called EEMD components) according to the EEMD approach. IMFs 1-5 represent each oscillation component with specific periods from high to low frequency. The trend (corresponding to the last IMF) is obtained after removing the oscillation components from the original time series. To examine which approach is better or has smaller error, we calculate the root-mean-square error (RMSE) between the derived time series by LSLF and EEMD and the raw original time series in observations and the CMIP5 models.

Table 1 Details of the Coupled Model Intercomparison Project Phase 5 (CMIP5) models used in this paper, including their modeling centers, countries (organizations), and references (or links) for the atmospheric component. Each model is labeled with a code number.

Both the LSLF and EEMD methods are effective tools and can be used for obtaining the climate trends from data (e.g., Qian et al., 2009, 2010; Franzke, 2012). Compared with the LSLF approach, the EEMD approach can extract the trend without requiring the time series to be linear and stationary (Wu et al., 2011; Qin et al., 2012). The trend deduced from the LSLF approach is highly sensitive to the choice of start and end points of the time series; the time intervals of a time series and the obtained trend in the past will change if the time series is extended (Wu et al., 2011), while the trend by the EEMD approach changes with time and the trend in the past will not change if the time series is extended. In this respect, the EEMD approach can derive a much more reliable trend evolution during a shorter time period, while the linear trend by the LSLF approach may be bias toward the time series at a particular position (Wu et al., 2007; Huang and Wu, 2008; Wu et al., 2011; Qin et al., 2012). To quantitatively compare the trend value with that by LSLF, we calculate the EEMD trend (or global warming rate) values by the approach used in Ji et al. (2014).

3　Results

The observed global averaged SATs from GISTEMP and HadCRUT4 reveal an obvious warming signal in the 20th century (1901-2005) with a positive linear trend of 0.345°C per five decade (Fig. 1). This warming signal iscomposed of a slow warming from 1901 to 1970, and a rapid warming (＞ 0.5°C) from 1971 to 2005. Most of the CMIP5 models capture the warming signal and its temporal features (Fig. 1), which is also presented in the time series of the (multi-model ensemble mean) MEM. The lines derived from LSLF fit the original time series for observations and simulations. However, the lines cannot capture well the upward trend in some models. The lines fitted by the EEMD trend with the addition of the multi-decadal variability (MDV) component can reproduce the original time series much better than the lines only derived from the EEMD trend component. During the period of 1901-2005, the trend values of the EEMD trend with the addition of the MDV component are larger than those of LSLF trends in most of the models (Table 2).

To examine the possible differences in the trend line during the slow and rapid warming periods by these two approaches, the fitted lines in these two periods are displayed in Fig. 2. During the slow warming period (1901-70), the lines by the EEMD trend component fit the original time series better than those by LSLF. Meanwhile, the trends by LSLF are larger than those by EEMD in observations and most of the models in this period (Table 2). During the rapid warming period (1971-2005), the lines by the EEMD trend component also fit the original time series better than those by LSLF, although the trends by LSLF are comparable with those by EEMD in observations and most of the models. In some models (FGOALS-g2, INM-CM4, CMCC-CM), the lines reconstructed by LSLF and EEMD fit very well,suggesting the trends are mainly linear in these models.

Table 2 Trend values (°C per five decade) calculated by LSLF and EEMD approaches. MEM is the multi-member ensemble mean.

Figure 3 shows the RMSEs between reconstructed lines (LSLF, EEMD trend component or EEMD trend + IMF5 components, IMF5 representing MDV component,～ 60 yr) and the original time series during 1901-2005. About half of the models have larger RMSEs between the reconstructed lines by the EEMD trend component and the original time series compared with that between the reconstructed lines by LSLF and the original time series. When the reconstructed lines include the EEMD MDV and trend components, smaller RMSE is obtained in most of the models compared with that by the reconstructed lines by LSLF during 1901-2005. This implies the MDV is important for estimating the temporally varying trend during the period of 1901-2005, which has been pointed out in some previous studies (e.g., Wu et al., 2011).

RMSEs are calculated during the slow and rapid warming periods by splitting the whole period (1901-2005) based on these two approaches. Smaller RMSEs are achieved in most of the models during these two shorter periods for the reconstructed lines only by the EEMD trend component by comparing those reconstructed straight lines for LSLF. Such a short length of time is not necessary to add the other component, except for the EEMD trend component to the reconstruction of the lines for fitting the original time series because the effect of MDV cannot be removed from the EEMD trend component (e.g., Ji et al., 2014). This also suggests it is better to estimate the trend using EEMD than LSLF especially for short time periods with obvious nonlinear behavior.

According to the above comparison, EEMD is a good choice to extract the trend (or global warming rate). In Table 2, the warming rates by the two approaches are given. During the period of 1901-2005, the trend values by EEMD trend + MDV components are larger (smaller) in 12 (10) models (total: 27) than those from observations. Among 12 (10) models, the larger (smaller) trend values also exist in 9 (8) models during the rapid warming period. This means for most of the models, if they have higher (lower) sensitivity to greenhouse gases (e.g., CO2), the trends (global warming rates) will be larger (smaller) during 1901-2005. During the slow warming period, 6 (5) models among 12 (10) models also have larger (smaller) trends than those from observations. This implies a high (low) sensitivity to greenhouse gases exists in the 6 (5) models for both the slow and rapid warming periods. This low and high sensitivity will lead to the smaller and larger global warming rates during the period of 1901-2005.

On a global scale, the EEMD trends are slightly lower than the LSLF trends over the slow warming period. The EEMD approach extracts the overall trend from the different time scales' internal variability. As is known, trends would be affected by sampling fluctuations of internal variability (Feldstein, 2002; Franzke, 2009), and the role of internal variability in global warming is potentially important, especially before the 1970s (e.g., Deser et al., 2012). Besides, the slight cooling from 1940-70 can be captured by the EEMD approach. Therefore, the EEMD trend can better reflect the slow warming before the 1970s because the LSLF approach cannot capture the internal variability at different time scales and in several cooling years. For an anthropogenic-caused trend during the rapid warming period, the EEMD approach is also demonstrated to be a powerful approach for extracting true trends.

4　Summary and conclusion

In this study, we use EEMD and LSLF to estimate the historical SAT trends in 27 CMIP5 models and observations from GISTEMP and HadCRUT4 during the period 1901-2005. We also examine the trends using these approaches during the slow and rapid warming periods. The magnitudes of the trends estimated by the two approaches are comparable. The trend calculated by the EEMD approach is larger than that by the LSLF approach in most of the models (23/27) during 1901-2005. The RMSEs between the raw time series and the reconstructed one by the LSLF approach are slightly smaller than that by the EEMD trend component, but larger than that by the EEMD trend and MDV components during 1901-2005 in most of the models and observations. This implies the MDV is important for estimating the temporally varying trend during the period 1901-2005. During the slow warming period, the EEMD trend is smaller than the LSLF trend. During 1901-70 (or 1971-2005), the RMSEs between the raw time series and the reconstructed one by LSLF are larger than those by the EEMD trend component. In this sense, the EEMD trend is a better choice to obtain the climate trends in observations and CMIP5 models, especially for short time periods. For short time periods (＜ 70 yr), the trend and MDV (～ 60 yr) are not clearly separated by EEMD IMFs. The LSLF approach cannot capture the SAT change during the slow warming period. This is because the LSLF cannot capture the internal variability for such short slow warming periods.

By comparing the trends by EEMD in different periods, we find the estimated global warming rates (trend) are larger (smaller) than those from observations during 1901-2005 for 11 of 27 models if the response to greenhouse gases (e.g., CO2) are higher (lower) during both the slow and rapid warming periods. EEMD can extract the signals at different time scales, and the trends. Using this approach, we can compare the trend and internal variability at different time scales and provide insight regarding the attribution and detection of global warming.

Acknowledgments. The authors gratefully acknowledge the helpful suggestions and comments from the two anonymous reviewers. We thank the modeling groups participating in CMIP5, and the Program for Climate Model Diagnosis and Intercomparison (PCMDI) for generously making the model output used in our report available to the community. This study is supported by the National Key Program for Developing Basic Sciences (Grant No. 2010CB950502), the National Natural Science Foundation of China (Grant Nos. 41376019 and 41023002), and the “Strategic Priority Research Program” of the Chinese Academy of Sciences (Grant No. XDA11010304).

References

Bao, Q., P. F. Lin, T. J. Zhou, et al., 2013: The flexible global ocean-atmosphere-land system model, spectral version 2: FGOALS-s2, Adv. Atmos. Sci., 30, 561-576.

Bentsen, M., I. Bethke, J. B. Debernard, et al., 2012: The Norwe-gian Earth System Model, NorESM1-M-Part 1: Description and basic evaluation, Geosci. Model Dev. Discuss., 5, 2843-2931.

Born, K., 1996: Tropospheric warming and changes in weather variability over the Northern Hemisphere during the period 1967-1991, Meteor. Atmos. Phys., 59(3-4), 201-215.

Chylek, P., J. Li, M. K. Dubey, et al., 2011: Observed and model simulated 20th century Arctic temperature variability: Canadian Earth System Model CanESM2, Atmos. Chem. Phys. Discuss., 11, 22893-22907.

Collins, W. J., N. Bellouin, M. Doutriaux-Boucher, et al., 2011: Development and evaluation of an Earth-system model—Had-GEM2, Geosci. Model Dev. Discussi., 4(2), 997-1062.

Dai, A., I. Y. Fung, and A. D. Del Genio, 1997: Surface observed global land precipitation variations during 1900-88, J. Climate, 10(11), 2943-2962.

Deser, C., A. Phillips, V. Bourdette, et al., 2012: Uncertainty in climate change projections: The role of internal variability, Climate Dyn., 38, 527-546.

Donner, L. J., B. L. Wyman, R. S. Hemler, et al., 2011: The dynamical core, physical parameterizations, and basic simulation characteristics of the atmospheric component AM3 of the GFDL global coupled model CM3, J. Climate, 24(13), 3484-3519.

Dufresne, J. L., M. A. Foujols, S. Denvil, et al., 2013: Climate change projections using the IPSL-CM5 Earth System Model: From CMIP3 to CMIP5, Climate Dyn., 40, 2123-2165.

Dunne, J. P., J. G. John, A. J. Adcroft, et al., 2012: GFDL's ESM2 global coupled climate-carbon earth system models. Part I: Physical formulation and baseline simulation characteristics, J. Climate, 25(19), 6646-6665.

Feldstein, S. B., 2002: The recent trend and variance increase of the annular mode, J. Climate, 15, 88-94.

Franzke, C., 2009: Multi-scale analysis of teleconnection indices: Climate noise and nonlinear trend analysis, Nonlinear Processes Geophys., 16, 65-76.

Franzke, C., 2010: Long-range dependence and climate noise characteristics of Antarctic temperature data, J. Climate, 23, 6074-6081.

Franzke, C., 2012: Nonlinear trends, long-range dependence, and climate noise properties of surface temperature, J. Climate, 25, 4172-4183

Gent, P. R., G. Danabasoglu, L. J. Donner, et al., 2011: The community climate system model version 4, J. Climate, 24(19), 4973-4991.

Hansen, J., R. Ruedy, J. Glascoe, et al., 1999: GISS analysis of surface temperature change, J. Geophys. Res., 104, 30997-31022.

Hazeleger, W., X. Wang, C. Severijns, et al., 2012: EC-Earth V2.2: Description and validation of a new seamless earth system prediction model, Climate Dyn., 39(11), 2611-2629.

Huang, N. E., and Z. Wu, 2008: A review on Hilbert-Huang transform: Method and its applications to geophysical studies, Rev. Geophys., 46(2), RG2006, doi:10.1029/2007RG000228.

Ji, F., Z. Wu, J. Huang, et al., 2014: Evolution of land surface air temperature trend, Nature Climate Change, in press, doi:10.1038/nclimate2223.

Jones, G. S., P. A. Stott, and N. Christidis, 2013: Attribution of observed historical near-surface temperature variations to anthropogenic and natural causes using CMIP5 simulations, J. Geophys. Res. Atmos., 118, 1-24.

Knutson, T., F. Zeng, and A. Wittenberg, 2013: Multi-model assessment of regional surface temperature trends: CMIP3 and CMIP5 20th century simulations, J. Climate, 26, 8709-8743, doi:10.1175/JCLI-D-12-00567.1.

Li, L., P. F. Lin, Y. Q. Yu, et al., 2013: The flexible global ocean-atmosphere-land system model, Grid-point Version 2: FGOALS-g2, Adv. Atmos. Sci., 30, 543-560.

Meehl, G. A., W. M. Washington, J. M. Arblaster, et al., 2013: Climate change projections in CESM1 (CAM5) compared to CCSM4, J. Climate, 26(17), 6287-6308.

Morice, C. P., J. J. Kennedy, N. A. Rayner, et al., 2012: Quantifying uncertainties in global and regional temperature change using an ensemble of observational estimates: The HadCRUT4 data set, J. Geophys. Res., 117, D08101, doi:10.1029/2011JD017187.

Norris, J. R., 2005: Trends in upper level cloud cover and surface divergence over the tropical Indo-Pacific Ocean between 1952 and 1997, J. Geophys. Res., 110(D21), doi:10.1029/2005JD-006183.

Qian, C., C. Fu, Z. Wu, et al., 2009: On the secular change of spring onset in Stockholm, Geophys. Res. Lett., 36, L12706, doi:10. 1029/2009GL038617.

Qian, C., Z. Wu, C. Fu, et al., 2010: On multi-timescale variability of temperature in China in modulated annual cycle reference frame, Adv. Atmos. Sci., 27, 1169-1182, doi:10.1007/s00376-009-9121-4.

Qin, Z., X. Zou, and F. Weng, 2012: Comparison between linear and nonlinear trends in NOAA-15 AMSU-A brightness temperatures during 1998-2010, Climate Dyn., 39, 1763-1779.

Raddatz, T. J., C. H. Reick, W. Knorr, et al., 2007: Will the tropical land biosphere dominate the climate-carbon cycle feedback during the twenty-first century? Climate Dyn., 29(6), 565-574.

Rotstayn, L. D., M. A. Collier, M. R. Dix, et al., 2010: Improved simulation of Australian climate and ENSO related rainfall variability in a global climate model with an interactive aerosol treatment, Int. J. Climatol., 30(7), 1067-1088.

Schmidt, G. A., M. Kelley, L. Nazarenko, et al., 2014: Configuration and assessment of the GISS ModelE2 contributions to the CMIP5 archive, J. Adv. Model. Earth Sys., 6, 141-184.

Stocker, T., D. Qin, G. K. Plattner, et al., 2013: Summary for policymakers, in: Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, Cambridge University Press, Cambridge and New York, 2-8.

Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design, Bull. Amer. Meteor. Soc., 93(4), 485-498.

Voldoire, A., E. Sanchez-Gomez, D. Salas y Mélia, et al., 2013: The CNRM-CM5.1 global climate model: Description and basic evaluation, Climate Dyn., 40, 2091-2121.

Volodin, E. M., N. A. Dianskii, and A. V. Gusev, 2010: Simulating present-day climate with the INMCM4.0 coupled model of the atmospheric and oceanic general circulations, Izvestiya, Atmos. Oceanic Phys., 46(4), 414-431.

Watanabe, M., T. Suzuki, R. O'ishi, et al., 2010: Improved climate simulation by MIROC5: Mean states, variability, and climate sensitivity, J. Climate, 23, 6312-6335.

Wu, Z., and N. E. Huang, 2009: Ensemble empirical mode decomposition: A noise-assisted data analysis method, Adv. Adaptive Data Anal., 1, 1-41.

Wu, Z., N. E. Huang, S. R. Long, et al., 2007: On the trend, detrending, and variability of nonlinear and nonstationary time series, PNAS, 104(38), 14889-14894.

Wu, Z., N. E. Huang, J. M. Wallace, et al., 2011: On the time-varying trend in global-mean surface temperature, Climate Dyn., 37, 759-773, doi:10.1007/s00382-011-1128-8.

Xin, X. G., T. W. Wu, J. L. Li, et al., 2013: How well does BCC_CSM1.1 reproduce the 20th century climate change over China? Atmos. Oceanic Sci. Lett., 6(1), 21-26.

Yukimoto, S., Y. Adachi, and M. Hosaka, 2012: A new global climate model of the Meteorological Research Institute: MRI-CGCM3: Model description and basic performance (special issue on recent development on climate models and future climate projections), J. Meteor. Soc. Japan, 90, 23-64.

Lin, P.-F., X.-L. Feng, and J.-J. Liu, 2015: Historical trends in surface air temperature estimated by ensemble empirical mode decomposition and least squares linear fitting, Atmos. Oceanic Sci. Lett., 8, 10-16,

10.3878/AOSL20140064.

17 July 2014; revised 18 August 2014; accepted 20 August 2014; published 16 January 2015

LIN Peng-Fei, linpf@mail.iap.ac.cn