Corresponding author: F. Drost, School of Earth Sciences, University of Melbourne, VIC 3100, Australia. (firstname.lastname@example.org)
 Previous studies have shown that various climate indices based on surface temperature can be used in detection and attribution studies of climate change. Besides global mean surface temperature, these indices are the contrast between surface temperature over land and over oceans, the temperature contrast between the Northern and Southern Hemispheres, the meridional temperature gradient in the Northern Hemisphere and the magnitude of the annual cycle of temperatures over land. The indices vary independently from the global mean at decadal timescales, yet show common responses to anthropogenic climate change. Collectively they are more useful in detecting and attributing climate change than global mean surface temperature alone. We use CMIP5 model data and investigate to what extent observed trends in surface temperature can be attributed to natural and anthropogenic forcings. The multi-model ensemble mean trend for all indices, except for NS, are either at or exceed the 5%–95% confidence interval for no trend. These trends cannot be explained by natural forcings only and additional forcings are required to replicate observed trends. Historical simulations with greenhouse gas forcings only resulted generally in trends in the indices that were larger than those in simulations with all historical forcings and observed. The difference in the trends in the indices between the simulations with all historical forcings and with greenhouse gas forcing only are ascribed to the effect of aerosols.
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.
 We use data from climate model simulations undertaken as part of the Coupled Model Intercomparison Project Phase 5 (CMIP5) [Taylor et al., 2012] with the aim of investigating to what extent observed trends in surface temperature since 1900 can be attributed to natural and anthropogenic forcings. We investigate surface temperature changes using five climate indices that describe the large-scale patterns of temperature variations [Braganza et al., 2003; Karoly and Braganza, 2001] and which have recently been updated using CMIP3 model data [Drost et al., 2011]. These climate indices, the land-ocean temperature contrast (LO), the meridional temperature gradient in the Northern Hemisphere (MTG), the magnitude of the annual cycle of temperatures over land (AC) and the inter-hemispheric temperature contrast (NS) contain information independent of the global mean surface temperature (GM) at decadal timescales [Braganza et al., 2003; Karoly and Braganza, 2001], and are therefore more useful in detecting and attributing climate change than global-mean surface temperature alone. In the aforementioned studies, it was found that observed and modelled trends in the indices over the last 50 years, except for NS, were significantly larger than expected from internal climate variability alone.
 Here we evaluate the climate response to various forcings used in the CMIP5 Detection and Attribution experiments over the period 1900–2005; all historical forcings (hist), historical greenhouse gas forcing only (histGHG) and historical natural forcings only (histNat). We use data only from those models that have submitted at least 3 simulations for the aforementioned scenarios to CMIP5 (see Table 1). Three observational datasets are used: the Goddard Institute for Space Studies land-ocean gridded temperature data (GISS) [Hansen et al., 2001, 2010], 1880–2010; the National Climatic Data Centre GHCN-ERSST (NCDC) [Smith and Reynolds, 2005; Smith et al., 2008], 1880–2010; and the Hadley Centre Climate Research Unit version 4 (HadCRUT4) [Jones et al., 2012; Morice et al., 2012], 1850–2010 datasets. As the result of incomplete coverage of observational data over various parts of the globe, a data mask based on the availability of observational data in the 20th century is applied to observations and model data [Drost et al., 2011].
An overview of the modelling centres, models, scenarios, the length of the control simulation (in years) and the number of simulations for hist, histGHG and histNat used in this analysis. For more information, see http://cmip-pcmdi.llnl.gov/cmip5/.
2. Decadal Variability
 Intrinsic variability at decadal timescales is determined from detrended data from all simulations and observations and is presented in Figure 1. With the exception of the GFDL-CM3 model, the standard deviation of global mean surface temperature in each model reflects observed decadal variability quite well. The variability for the three different forcings are similar to one another, with hist and histNat slightly higher due to the inclusion of solar and volcanic forcings, and are consistent with observed variability. The standard deviation of the other indices agree reasonably with observed, but are either slightly higher (LO, NS and MTG) or slightly lower (AC) than observed across most of the models. These comparisons are similar to those for the CMIP3 models [Drost et al., 2011], indicating no substantial change in model performance in simulating decadal temperature variability, apart from the GFDL-CM3 model which has substantially higher variability than observed.
3. Forced Responses
 Next we consider the temporal evolution and trends in each index and compare the CMIP5 model simulations to observations, to results using CMIP3 model data, and among the different forcings. For the same reason as in Drost et al. , the period of 1890–1920 is used as the base period. The 5%–95% confidence interval for zero trend is determined from the trends over 197 50 year periods, with an overlap of 25 years, taken from all control simulations listed in Table 1.
GM: The CMIP5 multi-model ensemble mean follows the observed global temperature very well (Figure 2). Comparing the results to that of the earlier study using CMIP3 data [Drost et al., 2011], the CMIP5 multi-model ensemble mean and the 5%–95% confidence range for GM are very similar to those of the CMIP3 analysis. The CMIP5 multi-model ensemble mean shows slightly less change than that of the CMIP3 multi-model ensemble mean after 1960, in better agreement with the observations. This is probably due to the absence of volcanic aerosol forcing in some CMIP3 simulations. The range of simulated twentieth century global mean temperature variations across the ensemble of CMIP5 simulations used in this study is similar to the CMIP3 models, due to the chaotic internal climate variability simulated in these models.
 Observed trends in global mean surface temperature over the last 50 years are well captured by the multi-model ensemble range of the CMIP5 models in hist, and they exceed the 5%–95% confidence interval for zero trend (Figure 3). The observed trends cannot be explained by natural variability (histNat) alone and are unlikely to be due to greenhouse gases (histGHG). The multi-model ensemble mean of histGHG results in a trend significantly larger than observed and simulated in hist, while the multi-model ensemble mean trend in histNat is not significant different from zero.
 The CMIP5 multi-model ensemble mean of GM in hist follows observed global mean surface temperature extremely well and clearly deviates away from that of histNat during the second half of the twentieth century (Figure 4). At the end of the twentieth century, the 5%–95% confidence interval for GM in hist and histNat are also clearly separated and GM in hist tends to follow GM in histGHG more. This result further indicates that the current trend in global mean surface temperature is not caused by natural forcings and that it is more consistent with the increase in greenhouse gas forcing. The difference in the trends between hist and histGHG can be attributed mainly to the cooling effect of anthropogenic aerosols in the all forcings simulations, as the main difference between hist and histGHG is the omission of anthropogenic aerosols in the histGHG simulations. When comparing the CMIP5 multi-model ensemble mean trends in hist and histGHG, the cooling effect of the anthropogenic aerosols and other non-GHG forcings here is equivalent to −0.83 ± 0.09 C/century over the last 50 years.
LO: The CMIP5 multi-model ensemble mean of LO in hist is lower than that determined for the CMIP3 data (Figure 2) over the last 50 years, and, unlike for GM, the multi-model ensemble range is more constrained in CMIP5 than in CMIP3, nearly halving the range determined for CMIP3. However, the multi-model ensemble range in hist for LO barely captures observed trends in LO (Figure 3). The individual model mean trends in LO for hist forcing are too low in general when compared to observed but they are still well separated from those determined for histNat. The multi-model ensemble mean trend in LO for histGHG is closer to observed, although there are significant variations among the individual model means. The fact that the CMIP5 multi-model ensemble mean trend in histGHG is closer to observed than the CMIP5 multi-model ensemble mean in hist is due to the fact that the trend in surface temperature over land in histGHG is much larger than in hist (2.93 ± 0.8 versus 1.82 ± 0.9 C/100 years) although they both compare favourable with observed temperature trends over land (2.33 ± 0.05 C/100 years). The differences between the rates in hist and histGHG can be ascribed to the large cooling effect of anthropogenic aerosols which predominantly affect land areas.
 Similar to GM, the CMIP5 multi-model ensemble mean follows observed LO quite well (Figure 4). The CMIP5 multi-model ensemble mean and the 5%–95% confidence interval for hist clearly deviates away from those of histNat, indicating that the trend in LO cannot be explained by natural forcings alone.
NS: The CMIP5 multi-model ensemble mean for NS in hist shows a larger trend over the last decade than the one determined using the CMIP3 data (Figure 2). However, the trend in the multi-model ensemble mean still does not exceed the 5%–95% confidence interval for zero trend (Figure 3 and compare with the findings in Drost et al. ). It has been hypothesized that the larger loading of anthropogenic aerosols in the Northern Hemisphere contributes to a relative cooling at mid to low tropospheric latitudes [Santer et al., 1996], thereby reducing the trend of a warming signal in NS. This theory is supported by the multi-model ensemble mean of NS in histGHG, which does not include aerosols, but does exceed the 5%–95% confidence interval for no trend.
MTG: The CMIP5 multi-model hist ensemble mean for MTG is higher than its equivalent for the CMIP3 data (Figure 2). The multi-model ensemble 5%–95% range is similar to that from CMIP3 but from the end of the twentieth century is above the zero-line. The multi-model ensemble range in hist captures observed trends in MTG well, and the multi-model ensemble mean trend, and the mean trend in the observations, exceed the 5%–95% confidence interval for zero trend. The trend in MTG in hist cannot be explained by natural forcings alone and could be explained by greenhouse forcing. As there is no significant difference in CMIP5 multi-model ensemble mean trends between hist and histGHG, one can conclude that the effect of anthropogenic aerosols on the trend in MTG does not differ much between the low and high northern hemisphere midlatitudes.
AC: Observed trends in AC have varied considerably over the last century. Since the middle of the twentieth century, the overall trend is negative, but this trend has been weakening over the last decade (Figure 2). The CMIP5 multi-model ensemble mean for AC in hist is similar to that of the CMIP3 analysis but its 5%–95% confidence interval has widened. The increase in the 5%–95% confidence interval can primarily be attributed to the fact that there is a large range in simulated trends among the models (Figure 3) and not to internal variability as simulated decadal variability in AC is relatively weak and possibly underestimated when compared to those in the other indices (Figure 1). The individual model mean trends in AC of only three models exceed the 5%–95% confidence interval for zero trend (Figure 3). However, the 95% confidence level for zero trend lies very close to the multi-model ensemble mean trend indicating that the trend in AC in hist among all models is nearly significant at the 90% confidence level. This trend cannot be attributed to natural forcings as the multi-model ensemble mean for AC in hist is either at, or just outside the 95% confidence level of histNat (Figure 4).
 Individual model mean trends in histGHG generally exceed the 5%–95% confidence interval for zero trend and hence one could conclude that the inclusion of anthropogenic aerosols in the all historical forcings simulations is the reason for the limited significance of the trend in AC. However one cannot exclude that changes in other forcings in hist (for instance land-use and ozone depletion) or modelled rainfall and cloud changes might be influencing trends in AC too. An example in question is the NCAR CCSM4 model. Although most models generally show a stronger negative trend in AC when excluding anthropogenic aerosols, the opposite is the case for the NCAR model. Whereas most models experience an increase in the trend for both Northern and Southern Hemisphere summer and winter temperatures when excluding aerosols, there is no such increase in the NCAR CCSM4 model for the Southern Hemisphere seasons and even a decrease in the trend for the Northern Hemisphere seasons when excluding aerosols, particularly during winter. This apparent impact of anthropogenic aerosols in the NCAR CCSM4 model is not consistent with simple physical considerations although it might to some extent explain why its trend in GM is larger than observed. The issues regarding the inclusion of aerosols in the CCSM4 model are well documented [Gent et al., 2011; Meehl et al., 2011] and it has been suggested that the omission of indirect effects of aerosols in CCSM4 might have led to simulated temperatures that are slightly too high.
 We have extended the analysis of five different indices of surface temperature changes using data from CMIP5 Detection and Attribution experiments: all historical forcings, greenhouse gas forcing only and natural forcings only for 1900–2005. The aim was to examine to what extent observed trends in surface temperature since 1900 can be attributed to natural and anthropogenic forcings.
 Observed decadal variability and trends were reasonably well captured by the models. Compared to the earlier analysis using CMIP3 data, the CMIP5 multi-model ensemble mean in GM and LO shows a smaller change during 1960–2000, in better agreement with observations, the CMIP5 multi-model ensemble mean trends in the climate indices NS and MTG have increased and the range of the CMIP5 multi-model ensemble in LO has become more constrained. Observational mean trends and the multi-model ensemble mean trends for all indices, except for NS, are either at or exceed the 5%–95% confidence interval for zero trend. For none of the indices could the observed trend be explained by natural forcings alone and additional forcing due to greenhouse gases is required to replicate the observed trends. Historical simulations with greenhouse gas forcings alone generally show trends in the indices that were larger than observed and than those in simulations with all historical forcings. The difference in the trends in the indices between the simulations with all historical forcings and greenhouse gas forcing only are mainly ascribed to the effect of anthropogenic aerosols in the all historical forcings simulations. The effect of aerosols on the global mean temperature was estimated as −0.83 ± 0.09 C/century over the last 50 years.
 We acknowledge the World Climate Research Programme's Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Table 1 of this paper) for producing and making available their model output. For CMIP, the U.S. Department of Energy's Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. This research was supported by funding from an Australian Research Council Federation fellowship (project FF0668679) and the Australian Research Council Centre of Excellence for Climate System Science (grant CE110001028). It was also supported by the NCI National Facility at the ANU via the provision of computing resources for the Earth System Grid.
 The Editor thanks the two anonymous reviewers.