Annular mode changes in the CMIP5 simulations

Authors


Corresponding author: N. P. Gillett, Canadian Centre for Climate Modelling and Analysis, University of Victoria, PO Box 3065, STN CSC, Victoria, BC, V8W 3V6, Canada. (Nathan.Gillett@ec.gc.ca)

Abstract

[1] We investigate simulated changes in the annular modes in historical and RCP 4.5 scenario simulations of 37 models from the fifth Coupled Model Intercomparison Project (CMIP5), a much larger ensemble of models than has previously been used to investigate annular mode trends, with improved resolution and forcings. The CMIP5 models on average simulate increases in the Northern Annular Mode (NAM) and Southern Annular Mode (SAM) in every season by 2100, and no CMIP5 model simulates a significant decrease in either the NAM or SAM in any season. No significant increase in the NAM or North Atlantic Oscillation (NAO) is simulated in response to volcanic aerosol, and no significant NAM or NAO response to solar irradiance variations is simulated. The CMIP5 models simulate a significant negative SAM response to volcanic aerosol in MAM and JJA, and a significant positive SAM response to solar irradiance variations in MAM, JJA and DJF.

1 Introduction

[2] In both hemispheres, sea level pressure (SLP) trends corresponding to decreases at high latitudes and increases in the subtropics have been observed [e.g., Marshall et al., 2004; Thompson et al., 2000]. These trends may be characterized by annular mode indices, which represent the strength of the SLP gradient at mid-latitudes in each hemisphere. These observed trends have the potential to affect regional temperatures and precipitation [e.g., Thompson et al., 2000], and in the Southern Hemisphere, to affect the circulation of the Southern Ocean, which in turn may influence the carbon cycle [e.g., Hall and Visbeck, 2002; Le Quéré et al., 2007]. For this reason, it is important to obtain reliable projections of changes in the annular modes.

[3] While some studies have long projected an increase in both the Northern Annular Mode (NAM) and Southern Annular Mode (SAM) indices in response to greenhouse gas increases [e.g., Fyfe et al., 1999; Gillett et al., 2002; Arblaster and Meehl, 2006; Marshall et al., 2004], the magnitude and even the sign of projected future changes in these indices remain subject to debate. For example, Son et al. [ 2008] and Perlwitz et al. [2008] both found that ozone recovery would drive a negative trend in the SAM index in austral summer in the coming decades, based on simulations with coupled chemistry models. In the Northern Hemisphere, early studies found NAM and North Atlantic Oscillation (NAO) trends of both signs in response to greenhouse gas increases [Gillett et al. 2002; Stephenson et al. 2006] and more recently, Morgenstern et al. [ 2010] found that a set of coupled chemistry-climate models projected a decrease in the NAM in DJF in response to increasing greenhouse gases. In particular, given the strongly negative NAO and NAM indices observed in Northern Hemisphere winters 2009-2010 and 2010-2011, some studies have suggested that reduced Arctic sea ice extent may render negative NAO events more likely, leading to a negative shift in the NAO in DJF [Jaiser et al. 2012; Francis and Vavrus, 2012]. The CMIP5 multi-model ensemble provides an unprecedented opportunity to test these hypotheses using a large ensemble of state-of-the-art climate models. Unlike the previous generation CMIP3 ensemble, all CMIP5 historical and scenario simulations include changes in stratospheric ozone, which is known to be an important driver of annular mode changes, particularly in the Southern Hemisphere [Thompson and Solomon, 2002; Gillett and Thompson, 2003]. Moreover, a standard dataset of stratospheric ozone change is prescribed in many CMIP5 models [Cionni et al., 2011], which includes an average of chemistry-climate model projections of ozone recovery through the 21st century, in contrast to the diverse set of data sets used in CMIP3, some of which omitted ozone recovery. Further, the number of models contributing to CMIP5 is much larger, and this study uses 37 separate coupled models for which the necessary data are available in the CMIP5 archive.

2 Data

[4] We use data from all CMIP5 models that have at least one historical simulation and one RCP 4.5 simulation, with sea level pressure data available for at least the period 1861–2099. Each historical simulation is merged with the corresponding RCP 4.5 simulation. This resulted in us using 37 CMIP5 models, and 75 individual simulations. Data from the following models were used: ACCESS1-0 (1), ACCESS1-3 (1), BCC-CSM1-1 (1), BCC-CSM1-1-M (1), BNU-ESM (1), CanESM2 (5), CCSM4 (6), CESM1-BGC (1), CESM1-CAM5 (3), CMCC-CM (1), CMCC-CMS (1), CNRM-CM5 (1), CSIRO-Mk3-6-0 (10), EC-EARTH (1), FGOALS-g2 (1), FGOALS-s2 (3), FIO-ESM (3), GFDL-CM3 (1), GFDL-ESM2G (1), GFDL-ESM2M (1), GISS-E2-H-CC (1), GISS-E2-R-CC (1), GISS-E2-R (6), HadGEM2-AO (1), HadGEM2-CC (1), HadGEM2-ES (1), INMCM4 (1), IPSL-CM5A-LR (4), IPSL-CM5A-MR (1), MIROC5 (3), MIROC-ESM-CHEM (1), MIROC-ESM (1), MPI-ESM-LR (3), MPI-ESM-MR (3), MRI-CGCM3 (1), NorESM1-ME (1), and NorESM1-M (1). Bracketed numbers indicate the number of individual ensemble members used from each model. A SAM index for each season was calculated following Gong and Wang [1999] by subtracting the zonal mean SLP at the model latitude closest to 65°S from the zonal mean SLP at the latitude closest to 40°S, derived from monthly mean model output. Results were not strongly sensitive to defining the index based on latitudes 5° further south. Similarly, a NAM index was calculated by subtracting zonal mean SLP at the model latitude closest to 65°N from zonal mean SLP at the model latitude closest to 35°N following Li and Wang [2003]. Similar results were obtained if the NAM and SAM indices were defined as the leading principal component of zonal mean sea level pressure anomalies poleward of 20° in each hemisphere (see Figures S1, S2 and S3 in the Supporting Information). An NAO index was calculated following Stephenson et al. [2006] by subtracting the mean SLP over a region north of 55°N, and between 90°W and 60°E from the mean SLP over a region between 20°N and 55°N and between 90°W and 60°E. We compare simulated circulation indices with indices derived from two spatially complete data sets derived from observations: A version of the gridded observational data set HadSLP2r [Allan and Ansell, 2006] with an update applied to post-2005 data to ensure continuity with data prior to this (HadSLP2r with reduced variance, available from http://www.metoffice.gov.uk/hadobs/hadslp2), and the 20th Century Reanalysis (20CR), into which only surface observations were assimilated [Compo et al., 2011]. Other reanalyses are more likely to have discontinuities in data-sparse regions associated with the introduction of satellite observations [e.g., Marshall, 2003].

3 Results

[5] Consistent with the CMIP3 models [Miller et al. 2006], the CMIP5 models on average show a progressive increase in the NAM, which is largest in DJF and SON (Figure 1). There is no evidence of a downward trend in the NAM index in DJF in response to greenhouse gas increases [e.g., Morgenstern et al., 2010; Jaiser et al., 2012], even though the rate of Arctic sea ice decline, which Jaiser et al. [2012] argue may drive a negative NAM trend, is larger in the CMIP5 models than in the CMIP3 models and more consistent with observations [Stroeve et al. 2012]. None of the individual CMIP5 models show a significant negative NAM response in DJF (Figure 2a) (we use a 10% significance threshold throughout this analysis). In the observed NAM index in DJF, a pronounced positive trend up to 2000 has been followed by a strongly negative NAM index in recent winters, but this is not reflected in the models, suggesting either that this is a manifestation of internal variability, or results from a mechanism not resolved by the models. Largely similar results are found for the NAO index in the CMIP5 models (Figure 1, top row). While a small positive trend is simulated in the multi-model ensemble mean NAO index in all seasons, the trend is largest in SON. To our knowledge, this has not been reported in other studies.

Figure 1.

A Stephenson et al. [2006] North Atlantic Oscillation (NAO) index (top), a Li and Wang [2003] Northern Annular Mode (NAM) index (middle) and a Gong and Wang [1999] Southern Annular Mode (SAM) index (bottom) are shown for the mean of 37 CMIP5 models’ merged historical and RCP 4.5 simulations (black) for each season. The gray band shows the range from the second to the 36th largest anomaly in each year based on a single ensemble member from each of the 37 models, a non-parametric estimate of the 5–95% confidence range. Colored lines show observational annular mode indices derived from HadSLP2r [Allan and Ansell, 2006] (red) and 20CR [Compo et al., 2011] (green) data. Simulated anomalies are shown relative to an 1861–1900 climatology, and observations are centered on the multi-model ensemble mean over the period for which they are shown.

Figure 2.

(a) NAM and (b) SAM changes between 1861–1910 and 2050–2099 for each season in the individual CMIP5 models and in the multi-model ensemble mean (bottom bars). Colored bars show 5–95% confidence ranges, derived for each model by first taking the ensemble mean over available ensemble members, calculating annular mode differences between 1861 and 2050, 1862 and 2051 and so on for each of the 50 years of each sample, taking the standard deviation over the resulting sample of 50 anomalies, dividing by the square root of 49 (the number of years minus one), and multiplying by the 5% cutoff value for a student-t distribution with 49 degrees of freedom. Figure 2b shows changes in the SAM in DJF between 1861–1910 and 2050–2099 in magenta and between 1980–2029 and 2050–2099 in gray. Dashed vertical lines show zero change for each season, and the tick intervals on the x-axes are 5 hPa.

[6] In the Southern Hemisphere, the simulated multi-model ensemble mean SAM anomaly over the period 1991–2010 is largest in DJF, which is due to a large contribution of stratospheric ozone depletion to the DJF change [Thompson and Solomon, 2002; Gillett and Thompson, 2003; Arblaster and Meehl, 2006] (the observed trend is largest in JJA, Figure S4). Perhaps for this reason, much attention has focused on the DJF SAM as a driver of changes in the Southern Ocean, including the Southern Ocean carbon cycle [e.g., Le Quéré et al., 2007]. However, a notable feature of the plot is that, by the end of the 21st century, by which time stratospheric ozone depletion is projected to have largely recovered, the SAM index is of a similar magnitude in all four seasons (Figure 1, Figure 2b), consistent with Perlwitz et al. [2008]. In DJF, the CMIP5 multi-model ensemble mean SAM index remains approximately constant over the 21st century under the RCP 4.5 scenario considered here, as the effects on the SAM of ozone recovery and ongoing greenhouse gas increases largely cancel each other. This is in contrast to some studies with coupled chemistry climate models, which find a decrease in the SAM index over this period [Son et al., 2008; Perlwitz et al., 2008]. None of the individual CMIP5 models simulate a significant decrease in the SAM index in DJF between 50 year periods centered on 2005 and 2075 under the RCP 4.5 scenario (Figure 2b). (If we compare the DJF SAM mean in the two 10 year periods 2090–2099 and 2000–2009 considered by Perlwitz et al. [2008], a single model (GFDL-CM3) simulates a significant negative SAM change).

[7] Previous observational and modelling studies have found a response of the NAM, NAO and SAM to volcanic aerosol [e.g., Miller et al., 2006; Karpechko et al., 2010]. After removing an 11-year running mean from the circulation indices, in order to remove variations associated with greenhouse gas changes, ozone changes and the solar cycle, we calculated mean NAO, NAM and SAM indices for seasons with an average aerosol optical depth greater than 0.07 [Sato et al. 1993]. Similar results were obtained after removing 21, 31, or 41 year running means from the indices. CMIP5 models in which volcanic aerosols were not specified were excluded from this analysis (CMCC-CM, CMCC-CMS, INMCM4, IPSL-CM5A-LR, and IPSL-CM5A-MR). Confidence in the multi-model ensemble mean response was assessed based on the spread across the CMIP5 ensemble, where we assume that the responses of individual models are independent. A comparison of the standard deviation of the volcanic responses in models from the same modelling center with the overall standard deviation indicates that this is a reasonable assumption. Unlike in the CMIP3 models [Miller et al. 2006], no positive NAM response is seen in the CMIP5 models in DJF (Figure 3a), consistent with previous results based on a smaller subset of CMIP5 models [Driscoll et al. 2012]. The DJF mean NAM response in the CMIP5 models is apparently significantly negative, although we note that this result is not robust to variations in the optical depth threshold, which change the seasons included in the composite, and thus we do not regard this as a robust result. Note that if we apply a similar analysis to the observations using the 20CR and HadSLP2r data sets, we do find a positive NAM and NAO response in DJF consistent with previous analyses, although the mean NAO and NAM anomalies are within the range of anomalies simulated in individual CMIP5 models, suggesting that the observed volcanic response is not statistically significant, at least based on the metric applied here.

Figure 3.

NAO, NAM and SAM responses to (a) volcanic aerosol and (b) solar irradiance variations. Figure 3a shows averages of the high-pass-filtered circulation indices for seasons in which the mean Sato et al. [1993] aerosol optical depth exceeded 0.07: SON 1883–MAM 1885 (Krakatoa), DJF 1902–JJA 1903 (Santa Maria), SON 1963–DJF 1963 and JJA 1964 (Agung), SON 1982–JJA 1983 (El Chichón) and SON 1991–DJF 1992 (Pinatubo). Figure 3b shows average regression coefficients of CMIP5 high-pass-filtered circulation indices onto the total solar irradiance used to force CMIP5 simulations [Frölich and Lean, 2004; Wang et al. 2005]. The 5–95% confidence ranges in the mean (colored bars) were derived by dividing the sample standard deviation across the available models by the square root of the number of models and multiplying by the 5% cutoff value of a student-t distribution with 31 degrees of freedom (a) or 36 degrees of freedom (b) (the number of models minus one). In all cases, results are based on the ensemble mean of each model.

[8] The positive SAM response to volcanic aerosol in SON, although consistent with Karpechko et al. [2010], is not statistically significant in the CMIP5 models (Figure 3a). Moreover, the response in MAM is significantly negative, in contrast to previous results based on two eruptions in the CMIP3 models [Karpechko et al. 2010], and the response in JJA is also significantly negative. These SAM results are robust to variations in the optical depth threshold, and are representative of the multi-model ensemble as a whole, rather than, for example, being driven by a strong response in one or two models. We note, however, that these are small responses compared to the interannual variability in the SAM (Figure 1).

[9] Several studies have found a response to solar irradiance variations in the NAM [e.g., Shindell et al., 2001] and in the SAM [e.g., Roscoe and Haigh, 2007]. Standard solar irradiance variations are specified in the historical simulations of all the CMIP5 models [Frölich and Lean, 2004; Wang et al. 2005]. In order to remove the response to multidecadal variations in greenhouse gases and ozone depletion, but not the solar cycle, we first removed a 21-year running mean from the circulation indices, and then regressed them onto the annual mean total solar irradiance [Frölich and Lean, 2004; Wang et al. 2005]. The regression was done separately for each model, and the mean response is shown in Figure 3b, along with its uncertainty range. No significant response to solar irradiance variations is seen in either the NAM or NAO indices. However, a positive SAM response to solar irradiance variations is seen in every season, and the response is significant in the multi-model ensemble mean in every season except SON (Figure 3b). While the response in the multi-model ensemble mean is significant, the range of regression coefficients in individual simulations includes zero, suggesting that this response is unlikely to be identifiable in observations, particularly also given that the observational record is shorter than the simulations considered here. Unlike the responses to greenhouse gases and stratospheric ozone, the amplitude of the response is small compared to the internal variability: a typical solar cycle peak-to-trough amplitude of ∼ 1 Wm  − 2 implies a SAM variation of only ∼ 0.2 hPa (Figure 3b).

[10] Most CMIP5 simulations include either specified or simulated variations in ozone associated with the solar cycle (V. Eyring et al., Long-term changes in tropospheric and stratospheric ozone and associated climate impacts in CMIP5 simulations, submitted to J. Geophys. Res., 2012). However, five of the models considered here did not include a representation of the solar cycle effect on ozone (CCSM4, IPSL-CM5A-LR, IPSL-CM5A-MR, MIROC5, MIROC-ESM). Exclusion of these models from our analysis of the solar response did not substantially affect our results, and these five models themselves showed an annual mean SAM response to solar forcing of a similar magnitude to that seen in the ensemble as a whole, although it was not statistically significant.

4 Discussion and Conclusions

[11] Future atmopheric circulation changes associated with the annular modes are expected to drive substantial changes in regional climate, the ocean circulation and carbon cycle. Here, we evaluate the annular mode projections of a much larger ensemble of climate models with more realistic forcings than the multi-model ensembles evaluated in previous such studies. On average, the CMIP5 models simulate an increase in the SAM, NAM and NAO indices in all four seasons between 1860 and 2100, with the largest increases simulated in the SAM. In the NAM, the increase is largest in SON and DJF and smallest in JJA, and in the NAO, the increase is largest in SON. In the SAM the increases simulated from 1860 to 2100 are similar in all seasons. The DJF SAM index shows a strong upward trend to 2000, and then remains approximately constant through the 21st century, likely due to the counteracting influence of greenhouse gas increases and ozone recovery. No individual CMIP5 model shows a significant decrease in the SAM over the 21st century in the RCP 4.5 scenario considered here, in contrast to some previous studies using other models and scenarios [Son et al., 2008; Perlwitz et al., 2008].

[12] No significant positive NAM or NAO response is seen in the CMIP5 models following volcanic eruptions. This could be due to missing or poorly simulated response mechanisms in the CMIP5 models, or may suggest that the positive mean NAM/NAO anomaly following eruptions in observations has arisen by chance in the small sample of post-volcanic years observed. No significant NAM or NAO response is seen in response to solar irradiance variations. Also, in contrast to previous studies [Karpechko et al., 2010], a significant negative SAM response is simulated in response to volcanic aerosol in MAM and JJA, although these responses are small in magnitude. The CMIP5 models also simulate a small significant positive SAM response to solar irradiance variations in MAM, JJA and DJF.

Ancillary