Uncertainties in future climate attributable to uncertainties in future Northern Annular Mode trend



[1] Atmospheric circulation variability associated with the Northern Annular Mode (NAM) modulates the climate over large areas in the Northern Hemisphere. Therefore, it is expected that future NAM changes in response to greenhouse gas concentration increases and other forcing will influence the climate change in these regions. Climate models simulate a wide range of future NAM changes, which introduces an uncertainty into regional climate predictions. To quantify this uncertainty we use the intermodel spread of the climate projections by the models participated in the Intergovernmental Panel on Climate Change Fourth Assessment Report. We show that the intermodel spread of the future NAM projections account for up to 40% of the variance of the surface temperature and precipitation projections over some regions in Eurasia and North America across the simulations. This result implies that the uncertainty in the future NAM makes a considerable contribution into the overall uncertainty in regional climate predictions.

1. Introduction

[2] Uncertainties in climate predictions arise from chaotic nature of climate variability, unknown future greenhouse gases (GHG) emissions, as well as from imperfections of the climate models. The uncertainty due to imperfections of the models may be assessed, for example, by comparing climate projections by several climate models forced by the same GHG emission scenario. Such studies are useful because they help to isolate processes responsible for the intermodel differences in climate projections. For example, poor representations of the cloud feedbacks [Dufresne and Bony, 2008] and the deep ocean heat uptake [Boé et al., 2009] are shown to be among the largest sources of uncertainty in future global temperature across the models used in the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4). The uncertainty due to internal climate variability may be assessed by comparing climate projections by the same climate model but initiated with different start conditions.

[3] On the regional scales additional sources of the uncertainty may also be important. Stott et al. [2006] showed that, on the continental scale, future warming rates are most uncertain in North America and Europe, regions whose climates are strongly influenced by the atmospheric circulation variability. In the Northern Hemisphere (NH) extratropics, the dominant mode of the circulation variability is the Northern Annular Mode (NAM) [Thompson and Wallace, 2000] whose positive (negative) phase corresponds to a poleward (equatorward) shift of the extratropical storm tracks. This variability has a pronounced impact on the NH climate [Hurrell, 1995; Thompson and Wallace, 2000]. Studies with climate models of different complexity suggest that the NAM is sensitive to GHG atmospheric concentration and will likely shift to a more positive phase as a response to future GHG emissions [Shindell et al., 1999; Fyfe et al., 1999; Gillett et al., 2003]. There is, however, a considerable spread in the future NAM projections across the climate models [Miller et al., 2006].

[4] Based on simulations with an atmosphere-only model, Sigmond and Scinocca [2010] suggested that the uncertainties in the future NAM represent a significant source of uncertainties in the future NH extratropical climate. It is however not clear, whether the NAM-related climate uncertainties are important when compared to other factors responsible for the intermodel spread of future climate projections assessed by Meehl et al. [2007]. The goal of this study is to quantify the contribution of the future NAM into the future climate uncertainty. We show that the intermodel spread in the future NAM does account for a considerable fraction of the intermodel spread in regional temperature and precipitation projections for the 21st century in the NH extratropics.

2. Data and Method

[5] We use data from the Coupled Model Intercomparison Project phase 3 (CMIP3) models used in the IPCC AR4. The data were retrieved from the World Climate Research Programme's (WCRP's) dataset at https://esgcet.llnl.gov:8443/. We concentrate on the SRES A1B scenario for which the largest number of simulations is available. For all available simulations monthly mean sea level pressure (SLP), surface air temperature, and precipitation were downloaded. Altogether data from 57 simulations for SLP and temperature and 56 simulations for precipitation by 24 models are used. Some models extend their SRES A1B simulations into the 22nd century, but only the 21st century data are used.

[6] We define the NAM as the first empirical orthogonal function (EOF) of monthly mean SLP north of 20°N weighted by the square root of cosine of latitude using the whole 21st century time series and all months. To avoid possible changes in EOF structure due to the long-term variability, linear trends were removed from the data at each grid point and for each month separately prior to the calculations. The NAM index was subsequently defined as a projection of original (not detrended) November–February averaged SLP time series on the NAM. The season was chosen based on previous studies [e.g., Shindell et al., 1999] which showed that the strongest NAM response to GHG increases is expected during winter. Influence of the NAM variability on temperature and precipitation is estimated using linear regression of the November–February averaged variables onto the NAM index at each grid point and allowing for autocorrelation as in the work of Gillett et al. [2006]. To avoid spurious regression due to solely common trends, the regression coefficients are calculated using detrended time series. To facilitate model comparison and averaging, temperature and precipitation data are interpolated onto a 5° × 5° grid prior to calculation, which is the approximate resolution of the coarsest model (GISS-E).

[7] The models employ different assumptions about the evolution of stratospheric ozone. Some models account for ozone depletion in the 20th century and, subsequently, prescribe ozone recovery in the 21st century. Others prescribe only annual variability that is constant from year to year. Although some studies suggest that the stratospheric ozone changes contributed to the NAM trend in the 20th century [Volodin and Galin, 1999] and will contribute in the 21st century [Paeth and Pollinger, 2010] we find that the difference between the future NAM trends in the two groups is not statistically significant (p > 0.1 according to a two-sample two-sided t-test) and therefore combine the models into one ensemble regardless of their stratospheric ozone forcing. Note that there are intermodel differences in other forcings such as tropospheric ozone and aerosols but these were not considered here.

[8] Future trends in the NAM index, temperature, and precipitation are calculated using linear regression of the seasonally averaged quantities on the time variable. The spread of the temperature and precipitation trend coefficients across the individual simulations is used as a measure of uncertainty in future climate. Similarly, the spread of NAM trend coefficients is used as a measure of uncertainty in the future NAM. A correlation between the climate variable trends and the NAM trends across the simulations is used to assess future climate uncertainty attributable to the future NAM uncertainty.

3. Results

[9] Figure 1 shows regression maps of temperature and precipitation on the NAM index averaged across the individual simulations. The anomalies shown correspond to the positive NAM phase. Figure 1 reveals features familiar from observations [Hurrell, 1995; Thompson and Wallace, 2000], which include positive temperature anomalies over northern Eurasia, negative temperature anomalies over eastern Canada and western Greenland, positive precipitation anomalies over the North Atlantic and the Northern Europe, and negative precipitation anomalies over the subtropical Atlantic and the Mediterranean. These anomalies are attributable to anomalous heat advection and shifts of the storm tracks associated with the NAM variability, and are discussed in detail elsewhere [Hurrell, 1995; Thompson and Wallace, 2000]. Figure 1 suggests that the models are able to simulate the climate impacts of the NAM reasonably well, in agreement with previous studies based on an earlier climate model generation [e.g., Stephenson et al., 2006].

Figure 1.

Multi-model mean regression of November–February (a) surface temperature (K/hPa) and (b) precipitation (mm/day/hPa) on the November–February NAM index. The anomalies correspond to the positive NAM phase. Dashed contours are for negative values. Hatching denotes areas where 95% of the simulations agree on the sign of regression. Thick black lines indicate areas used to calculate the trends shown in Figure 2.

[10] We expect that the future NAM trend will have a strong impact on climate change in the regions influenced by the NAM variability. As an example, Figure 2 shows scatterplots of the 21st century trends in temperature and precipitation across individual simulations for the two regions in central Siberia (temperature) and in the vicinity of the Iberian Peninsula (precipitation) plotted against the corresponding NAM trends. In these regions the models simulate warm (Siberia) and dry (the Iberian Peninsula) anomalies during the positive NAM phase (Figure 1). Figure 2 shows that the simulated climate change in these regions depends significantly on the simulated NAM trends: models with larger NAM trends simulate a stronger warming in Siberia and a larger reduction in precipitation in the Iberian Peninsula. The dependence emerges regardless of considerable intermodel differences in the climate feedbacks and the rates of heat uptake by the ocean reported in other studies [Soden and Held, 2006; Dufresne and Bony, 2008; Boé et al., 2009]. In both regions, the spread of the future NAM trends explains ∼40% of the variance of the future climate trends across the individual simulations.

Figure 2.

21st century trends in November–February (a) surface temperature (K/decade) and (b) precipitation (mm/day/decade) across the CMIP3 simulations plotted against the trends in the November–February NAM index. The temperature and precipitation trends are calculated for the areas indicated in Figure 1. Different symbols correspond to different models. Black solid lines indicate linear fits calculated using all simulations. Also shown are linear fits for the models that have four or more simulations: MPI ECHAM5 (solid red), NCAR CCSM3 (dashed red), NCAR PCM1 (dotted red), GISS-ER (solid yellow), CCCMa CGCM3.1 T47 (solid blue), and MRI CGCM2.3.2a (solid green).

[11] A part of the spread in the future NAM trends shown in Figure 2 is attributable to the internal climate variability. This is suggested by the different NAM trends across multiple climate simulations by the same models. These simulations differ only in their initial conditions; therefore these different NAM trends cannot be attributed to different model sensitivity to the GHG concentrations. Rather, they arise due to the internal climate variability. Figure 2 shows that the dependence of the future climate trends on the future NAM trends across multiple simulations by the same model is similar to that across the all simulations: larger NAM trends correspond to warmer (dryer) future conditions in these regions. One can estimate the relative contributions of the internal climate variability and the intermodel differences to the total spread of the future NAM trends by applying the analysis of variance (ANOVA) technique [e.g., von Storch and Zwiers, 1999]. The results show that the part of the variance due to the intermodel differences is larger than that due to the internal climate variability by a factor of 3. Although the contribution of the internal climate variability is considerable, this result suggests that the intermodel differences, not the internal climate variability, are the major contributor to the spread of the future NAM trends across the CMIP3 models. This implies that climate prediction uncertainties in these regions may be reduced if the reasons for the intermodel spread in the NAM trends are understood and possible model biases are corrected.

[12] The fraction of the variance of future temperature and precipitation trends explained by the spread of the future NAM trends across the individual simulations (i.e. squared Pearson's correlation coefficient between the NAM trends and the climate variables trends) is shown in Figure 3 for each grid point north of 20°N. As expected, the regions of significant NAM influence on the future climate trends largely coincide with the regions influenced by the NAM variability (Figure 1). A large influence of the simulated future NAM on the temperature trends is found in central Eurasia, Japan, North Pacific, and southeast North America. The largest influence on the precipitation trends is in the Mediterranean, the Northern Europe, and northwest Siberia. In these regions the NAM-explained variance exceeds 40% for the temperature trends and 20–30% for the precipitation trends. This result suggests that the uncertainties in future NAM make a considerable contribution into the overall uncertainties in the future climate in these regions. The NAM uncertainty contributes to the future warming rate uncertainties in Europe and North America; however its influence there is limited, especially in Europe, and is unlikely to explain the large uncertainties reported by Stott et al. [2006].

Figure 3.

The fraction of the intermodel variance of the 21st century November–February (a) surface temperature and (b) precipitation trends that is explained by the intermodel spread in the November–February NAM index trends. Hatching denotes areas where the correlation between the NAM trends and the climate variable trends is significant at the 1% level.

[13] A closer examination of Figure 3 reveals also noticeable deviations from the patterns shown in Figure 1. For example the fraction of uncertainties in future temperatures attributable to the uncertainties in future NAM is small in the eastern Canada and western Greenland region, despite the fact that the models simulate a significant temperature response to the NAM variability there. In this region, the spread of temperature trends across the CMIP3 simulations exhibits a local maximum (not shown). We speculate that the factors responsible for this spread are most likely related to the simulations of sea ice response to GHG increases. In this region, the warming is affected by the retreat of sea ice and, thus, depends on the simulation of the present day climate sea ice edge. Chapman and Walsh [2007] showed that the models exhibit a large spread in their simulations of the present day climate sea ice edge, which consequently leads to a large spread of the simulated warming in this region. As a result, changes in the atmospheric heat advection related to the changes in the NAM account for only a small fraction of the total heat budget changes in this region. The dependence of the temperature trends on the NAM trends in this region is, however, seen in the multiple simulations by the same model (not shown) further suggesting that it is the intermodel differences that obscure the NAM-temperature relationship within the multi-model ensemble in this region. A significant (though small) correlation between the NAM trends and the climate variables trends is also found in some regions not influenced by the NAM variability and even in the Southern Hemisphere, which may be related to the same model biases that control the NAM sensitivity to GHG forcing. Not surprisingly, it is not accompanied by similar correlations across the multiple simulations by the same models.

4. Conclusions and Discussion

[14] We have demonstrated that the intermodel spread of the simulated future NAM projections accounts for a considerable fraction of the intermodel spread of the future temperature and precipitation projections across the CMIP3 models in some regions. These regions are largely, but not exactly, those influenced by the NAM variability. We also found that, in some regions influenced by the NAM variability, other intermodel differences play a more important role in producing the spread in the future climate projections. These results help to understand sources of uncertainties in future climate. They also highlight the importance of understanding and properly accounting for in the climate models the mechanisms controlling the NAM response to GHG emissions. Presently these mechanisms are poorly understood, and are subject of ongoing research.

[15] Based on the fluctuation-dissipation theory, Ring and Plumb [2008] suggested that the NAM response may depend on the decorrelation timescale of unforced NAM anomalies. This theory links the deccorelation timescale of unforced anomalies of an atmospheric mode to the response of this mode to an external forcing. Kidston and Gerber [2010] have found a correlation between the deccorelation timescale, position of the extratropical jet stream in the 20th century, and the GHG response across the CMIP3 models for the Southern Annular Mode. Although it is not clear what fraction of the spread in the NAM projections can be explained by this dependence, or what controls the decorrelation timescales, the results of Kidston and Gerber [2010] suggest that a correct simulation of the present day extratropical circulation is prerequisite for a correct simulation of the NAM response to GHG forcing.

[16] Some of the biases in the simulated tropospheric circulation may be related to biases in the stratospheric circulation [Gerber et al., 2010]; therefore improvement of the representation of the stratosphere in the models may be a way to improve the simulations of the NAM response. Recently, Sigmond et al. [2008] and Sigmond and Scinocca [2010] demonstrated that, in their model, the NAM response to a doubling CO2 differ significantly between simulations with a unrealistic stratospheric zonal circulation (i.e. with a degraded stratosphere) and simulations with a properly represented stratospheric circulation. A step further would be to understand what part of the spread of the NAM projections shown in Figure 2 is attributable to the intermodel differences in the present climate stratospheric circulation. One way to do it is to repeat the IPCC scenario runs using two versions of the same model which differ in their stratospheric part. This work is under way.


[17] A.K. was funded by Finnish Academy under post-doctoral grant. We thank Nathan Gillett and two anonymous reviewers for their useful comments and the modeling groups, the Program for Climate Model Diagnosis and Intercomparison (PCMDI), and the WCRP's Working Group on Coupled Modelling (WGCM) for their roles in making available the WCRP CMIP3 multi-model dataset. Support of this dataset is provided by the Office of Science, U.S. Department of Energy.