Attribution of observed sea level pressure trends to greenhouse gas, aerosol, and ozone changes
Nathan P. Gillett,
Canadian Centre for Climate Modelling and Analysis, Environment Canada, Victoria, British Columbia, Canada
Corresponding author: N. P. Gillett, Canadian Centre for Climate Modelling and Analysis, Environment Canada, University of Victoria, PO 1700, STN CSC, Victoria, BC, V8W 3 V6, Canada. (email@example.com)
The copyright line for this article was changed on 13 APR 2015 after original online publication.
 Human influence on atmospheric sea level pressure (SLP) has previously been detected globally, but the contributions of greenhouse gas, aerosol, and ozone changes to the observed trends have not been separately identified. We use simulations from eight climate models to show that greenhouse gas, aerosol, and ozone changes each drive distinct seasonal and geographical patterns of trends, which are separately detectable in observed seasonal SLP trends over the 1951–2011 period. This detection is driven by significant low-latitude SLP responses to greenhouse gas, aerosol, and ozone changes, as well as the more frequently-studied high latitude responses. These results aid in understanding past atmospheric circulation changes, and have potential to improve projections of future circulation changes.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Changes in sea level pressure and the associated atmospheric circulation changes have driven pronounced regional changes in precipitation, surface temperature, ocean circulation, and the carbon cycle [e.g., Shindell et al., 2001; Thompson and Solomon, 2002; Gillett and Thompson, 2003; Kang et al., 2011; Le Quéré et al., 2007; Sigmond et al., 2011]. It is therefore important to quantify the contributions of different climate forcings to these changes. Several studies have detected human influence in atmospheric sea level pressure globally [Gillett et al., 2003, 2005; Gillett and Stott, 2009], but these studies did not separate the influence of individual climate forcings. Modeling studies suggest that changes in greenhouse gases [Fyfe et al., 1999; Shindell et al., 1999, 2001; Marshall et al., 2004; Arblaster and Meehl, 2006; Sigmond et al., 2011], aerosols [Rotstayn et al., 2007; Bollasina et al., 2011; Ming and Ramaswamy, 2011; Ming et al., 2011; Allen et al., 2012], and stratospheric ozone [Shindell et al., 2001; Gillett and Thompson, 2003; Marshall et al., 2004; Arblaster and Meehl, 2006; Sigmond et al., 2011] have contributed to these changes, but differ in the relative importance they ascribe to each forcing. Here we identify and quantify the contributions of greenhouse gases, ozone, and aerosol to observed sea level pressure trends, using a set of climate model simulations and a detection and attribution analysis.
 We start by comparing patterns of simulated and observed SLP trends over the period 1951–2011 during which observational coverage has been relatively good. We derive observed trends for each season from HadSLP2 [Allan and Ansell, 2006] updated from 2005 with HadSLP2r_lowvar (see supporting information) and compare these with simulations from the Fifth Coupled Model Intercomparison Project (CMIP5; Table S1) of the response to greenhouse gas (GHG), aerosol (AER), and stratospheric, and tropospheric ozone [Eyring et al., 2013] (OZ) changes (see supporting information). While we expect the stratospheric component of ozone change to have been dominant in driving Southern Hemisphere (SH) circulation changes [Gillett and Thompson, 2003], tropospheric ozone changes may have contributed to simulated circulation changes in the Northern Hemisphere (NH) [Allen et al., 2012]. Aerosol simulations include sulphate, black carbon, and organic carbon aerosol changes. In order to focus attention on the regions in which the trends are largest compared to internal variability, we plot the ratio of the observed or simulated trend in each grid cell to the standard deviation of control simulation trends (Figure S1). The internal variability of SLP trends is much greater at high latitudes than at low latitudes, so this approach emphasizes low latitude trends.
 The SLP trends simulated in response to GHG changes (Figure 1) exhibit an increase over the SH midlatitudes and a decrease over the SH high latitudes, which project onto the positive phase of the Southern Annular Mode (SAM) [e.g., Fyfe et al., 1999; Marshall et al., 2004; Arblaster and Meehl, 2006] (Figure 2b), and a weaker high latitude response in the NH, which projects onto the positive phase of the Northern Annular Mode (NAM) [e.g., Fyfe et al., 1999] (Figure 2a). However, broad regions of significant change are also simulated over the low latitudes, including SLP increases over Southeast Asia and much of the western and central tropical Pacific, largest in December–February (DJF) but persisting through most of the year, as well as weaker decreases in SLP over the eastern tropical Pacific. This pattern corresponds to a weakening of the Walker circulation [Vecchi et al., 2006]. Consistent with previous results [Bollasina et al., 2011], significant SLP increases are also simulated in June–August (JJA) over much of the tropical Indian Ocean, with decreases over the Middle East, Europe, and much of Africa.
 The ozone response exhibits a well-known decrease in SLP in DJF over the SH high latitudes and an increase in the SH midlatitudes (Figure 1), corresponding to an increase in the SAM index [e.g., Thompson and Solomon, 2002; Gillett and Thompson, 2003; Sigmond et al., 2011] (Figure 2b). However, ozone changes also drive a significant increase in SLP over much of the tropical Pacific in DJF. Impacts of ozone changes on low-latitude precipitation have previously been identified [Kang et al., 2011], but this effect on SLP in the tropical Pacific has not previously been noted, and the mechanism remains to be identified. In March–May (MAM) (Figure S2) there is a significant increase in SLP in the NH subtropics, and a weaker decrease further north, corresponding to an increase in the NAM (Figure 2a), which may reflect a response to tropospheric ozone changes [Allen et al., 2012]. Simulations from CanESM2 including only stratospheric ozone changes did not exhibit this upward trend in the NAM in boreal spring, while those also including tropospheric ozone changes did (Figure S3a).
 The SLP response to aerosols is weaker than the response to GHGs or ozone, but nonetheless it does contain some significant features. The response is largest in JJA when the radiative forcing associated with the aerosol, which is mainly emitted in the NH, is largest (Figure 1). SLP decreases are simulated over most of Europe and the North Atlantic, while significant increases are simulated over South Asia [Rotstayn et al., 2007; Bollasina et al., 2011]. There is a general decrease in SLP at ~30°S [Rotstayn et al., 2007; Bollasina et al., 2011; Ming and Ramaswamy, 2011], which is largest in DJF and corresponds to a negative trend in the SAM [Arblaster and Meehl, 2006] (Figure 2b). The response to natural (solar and volcanic) forcings (NAT) contains few regions of significant trends (Figure S2), as might be expected since volcanic aerosol and solar irradiance exhibit only small trends over this period.
 Some similarities between the observed SLP trends and the simulated responses to forcings are readily apparent (Figure 1). For example, consistent with the GHG response, increases in SLP are observed in JJA over Southeast Asia and the Indian Ocean, and decreases over North Africa. Consistent with the AER response, SLP increases are observed in JJA over India and decreases observed over much of Europe and the North Atlantic. Consistent with the OZ response, a large SAM-like response is observed in DJF. Nonetheless, at many individual locations, observed trends are relatively small compared to internal variability, and the NAM trend is not generally statistically significant over this period (Figure 2a). In other regions, observations and simulations disagree, such as in the eastern tropical Pacific, where the models simulate a decrease in SLP [Vecchi et al., 2006], but an increase has been observed in most seasons over the 1951–2011 period (Figure 1). Overall trends simulated in response to combined anthropogenic and natural (ALL) forcings are consistent at the 10% level with observed trends to within internal variability in 71% of grid cells. This indicates good, but not perfect, agreement between simulated and observed grid cell SLP trends. One possible cause of this small discrepancy is the direct effect of changes in column water vapor on sea level pressure [e.g., Trenberth and Smith, 2005], which is not accounted for in some models.
 In order to properly account for the role of internal variability and the superposition of the responses to multiple forcings, we apply a detection and attribution analysis to determine the causes of the observed trends. We start by using output from six CMIP5 models to attempt to detect the responses to GHG, NAT, and a combined other anthropogenic response (OTH), dominated by ozone and aerosol changes, since the simulations necessary to separate the aerosol and ozone responses were only available from two models (Table S1). We apply a total least squares regression [Allen and Stott, 2003] to the SLP trends in all four seasons simultaneously, and project on the leading 60 Empirical Orthogonal Functions of control variability (see supporting information). The first six sets of bars in Figure 3a show regression coefficients of observed seasonal trends against GHG, OTH, and NAT responses, derived using the forced response patterns from each of the six models separately. The GHG regression coefficients were inconsistent with zero, and hence the GHG response was detected, using five of the six models, the OTH response was detected using four of the models, and the NAT response was detected using two of the models. We note that only small ensembles of the response to each forcing were available from some models (Table S1), inflating uncertainties in the individual model analyses. Some individual models have regression coefficients significantly less than one indicating that they over-predict the SLP response to GHGs and to combined ozone and aerosols (BCC-CSM1-1, GISS-E2-H, and GISS-E2-R) (Figure 3a). When response patterns were averaged together across models [e.g., Gillett et al., 2002], GHG and OTH responses were robustly detected (Figure 3a), and model mean responses were consistent in magnitude with observations.
 To attempt to separate the responses to aerosols and ozone changes, we restricted our attention to the two models with aerosol-only simulations to 2011, as well as suitable combined anthropogenic and natural (ALL), GHG, and NAT simulations: CanESM2 and CSIRO-Mk3-6-0. An attribution analysis indicated detectable GHG, AER, and OZ signals using either of the two models (first two sets of bars in Figure 3b). A multi-model analysis combining output from CanESM2 and CSIRO-Mk3-6-0 indicated detectable GHG, AER, and OZ signals (rightmost set of bars in Figure 3b), of consistent magnitude in simulations and observations. Overall we find that averaged over all seasons and observed locations (Figure S2), GHG, OZ, and AER have made comparable contributions to observed SLP trends.
 We find that greenhouse gas, aerosol, and ozone changes have each made distinct significant contributions to observed SLP trends over the past 60 years, over low latitudes as well as high latitudes. While a combined response to external forcing has previously been detected in SLP [Gillett et al., 2003, 2005], and a combined anthropogenic response has been detected independently of a natural response [Gillett and Stott, 2009], this is the first time that greenhouse gas, ozone, and aerosol influences have been separately detected. In a multi-model analysis, regression coefficients of observed SLP trends against the simulated responses to forcings are found to be consistent with one, indicating that, on average, the CMIP5 models are able to simulate the SLP responses to these forcings realistically. The SLP responses to greenhouse gases, aerosols, and ozone have distinct zonal, meridional, and seasonal structures, which aid the separation of their influences in a detection and attribution analysis. Previous studies using only zonal means [Gillett and Stott, 2009], or trends in one season [Gillett et al., 2003, 2005], therefore missed some of this structure, limiting their ability to separate the influences of individual forcing components. The signal-to-noise ratio is also enhanced in our analysis by our use of multiple models, and by our use of observations and simulations to 2011. These forced SLP trends have almost certainly influenced wind patterns, temperatures, and precipitation around the globe [Thompson and Solomon, 2002; Gillett and Thompson, 2003], as well as the ocean circulation and carbon cycle in the Southern Ocean [Le Quéré et al., 2007; Sigmond et al., 2011]. Through the 21st century, as greenhouse gas concentrations increase, aerosol precursor emissions decline and stratospheric ozone recovers, we can expect an evolving pattern of ongoing forced SLP changes and associated impacts across the globe, which is distinct from the historical trend.
 We thank Gareth Marshall (British Antarctic Survey) for providing station sea level pressure data. D.E.P. was supported by the Joint DECC/Defra Met Office Hadley Centre Climate Programme (GA01101), UK. Support for the Twentieth Century Reanalysis Project was provided by the U.S. Department of Energy, Office of Science Innovative and Novel Computational Impact on Theory and Experiment (DOE INCITE) program, and Office of Biological and Environmental Research (BER), and by the National Oceanic and Atmospheric Administration Climate Program Office. N.P.G. carried out the analysis and wrote the paper. J.C.F. provided advice and input on the analysis and interpretation of the results. D.E.P. developed the HadSLP2r_lowvar dataset used in this analysis.
 The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.