Current consensus on global climate change predicts warming trends driven by anthropogenic forcing, with maximum temperature changes projected in the Northern Hemisphere (NH) high latitudes during winter. Yet, global temperature trends show little warming over the most recent decade or so. For longer time periods appropriate to the assessment of trends, however, global temperatures have experienced significant warming trends for all seasons except winter, when cooling trends exist instead across large stretches of eastern North America and northern Eurasia. Hence, the most recent lapse in global warming is a seasonal phenomenon, prevalent only in boreal winter. Additionally, we show that the largest regional contributor to global temperature trends over the past two decades is land surface temperatures in the NH extratropics. Therefore, proposed mechanisms explaining the fluctuations in global annual temperatures should address this apparent seasonal asymmetry.
 Global surface temperatures are projected to warm due to rapid increases in greenhouse gases (GHGs) and over the entire length of the instrumental record, temperatures have undergone a warming trend [Intergovernmental Panel on Climate Change, 2007]. However it has been noted that the global warming trend has slowed and even halted since the late 1990s [e.g., Easterling and Wehner, 2009; Kaufmann et al., 2011]. Several studies have offered explanations for the recent lapse in the global temperature trend, including climate variations of the North Atlantic and equatorial Pacific oceans [Keenlyside et al., 2008; Meehl et al., 2011], variations in solar activity [Lean and Rind, 2009], changes in stratospheric water vapor [Solomon et al., 2010], and tropospheric aerosols [Kaufmann et al., 2011]. The sudden absence of a warming trend is not particularly unique or even unexpected. Easterling and Wehner , through running decadal trend analysis, show that even the observed record has decadal spans of little to no warming. Even coupled climate model simulations of a future warmer world due to increased GHGs contain periods of little to no warming [Easterling and Wehner, 2009; Meehl et al., 2011]. Indeed, large-scale modes of natural variability in the atmosphere and ocean may work to negate radiative warming on decadal timescales.
 The aforementioned studies discuss the absence of a warming trend or even a cooling trend in the context of global annual temperatures. Yet, analysis of temperature trends seasonally is not extensively reported, both in the observational record and in coupled climate model simulations. As we will illustrate, recent trends in global surface temperatures are seasonally-dependent; i.e., only Northern Hemisphere (NH) extratropical land surface temperatures during boreal winter display a systematic waning of the warming trend to a near-neutral trend of late. The implications of this finding are important for testing the relative importance of the proposed mechanisms suggested for the cessation of the warming trend and for evaluating how coupled climate models can handle such seasonal asymmetries.
2. Observational Data and Model Output
 Observational surface temperature data originate from two sources: (1) The Climate Research Unit land air temperature dataset, version 3 (CRUTEM3) [Brohan et al., 2006]; and (2) The National Aeronautics and Space Administration Modern Era Retrospective-Analysis for Research and Applications (NASA MERRA) [Rienecker et al., 2011]. The CRUTEM3 dataset, consisting of monthly land station-based temperature anomalies, are on a regular 5° × 5° longitude/latitude grid globally. The data are provided as anomalies, using the 1961–1990 period as the base period, and extend from 1850 to present. NASA MERRA monthly-mean surface temperatures reside on a 1.25° × 1.25° grid globally, covering land and ocean, starting from 1979 to present. Monthly-mean anomalies from MERRA are computed by removing the climatological monthly means (1979–2011) from the raw data.
 Coupled-climate model output are provided from the Coupled Model Intercomparison Project Phase 5 (CMIP5) multi-model ensemble archive, available for download from the Program for Climate Model Diagnosis and Intercomparison (PCMDI) at the Lawrence Livermore National Laboratory (more information on the program is provided online athttp://cmip-pcmdi.llnl.gov/cmip5/). We used all available ensemble members of the decadal1980 scenario (i.e., the model runs are initialized with conditions as in 1980 and run for 30 years using mainly observed natural and anthropogenic forcing). Six models were available and analyzed for this study (see Table 1), each with multiple realizations. Surface temperature output from the models is interpolated from their native model grids to a common 2.5° by 2.5° longitude/latitude grid for comparison. Surface temperature anomalies are then computed by removing the climatological monthly mean from 1980–2010. Model results are shown as the ensemble-mean of all models and all realizations.
Table 1. The Six Coupled Climate Models From the CMIP5 Model Archive Analyzed for This Studya
Modelling Agency, Country
Number of ensemble members for the decadal1980 scenario is also indicated.
Canadian Climate Centre for Modelling and Analysis, Canada
Météo-France/Centre National de Recherches Météorologiques, France
Hadley Centre for Climate Prediction and Research/Met Office, United Kingdom
Center for Climate System Research, Japan
Center for Climate System Research, Japan
Meteorological Research Institute, Japan
 When examining seasonal temperature trends, seasons are defined as follows: December–February (DJF) as boreal winter, March–May (MAM) as boreal spring, June–August (JJA) as boreal summer, and September–November (SON) as boreal fall. To test the robustness of any trends, given that the period 1999–2010 spans twelve years, we computed and tested the significance of multiple periods divisible by six. Therefore we also computed the temperature trend for the following periods: 2005–2010, 1999–2010, 1993–2010 and 1987–2010. For completeness, we also computed the trend and significance for the longer period 1979–2010, which includes the full period that satellite data is assimilated into the reanalysis datasets [Kalnay et al., 1996]. Trends for all periods are shown in degrees Celsius per five years for comparative purposes.
3. Results and Discussion
Figure 1 presents the linear temperature changes from the CRUTEM3 dataset over five periods (1979–2010, 1987–2010, 1993–2010, 1998–2010, and 2005–2010) and four regions: globally (Figure 1a), the NH extratropics (20–90°N; Figure 1b), the tropics (20°N–20°S; Figure 1c) and the Southern Hemisphere (SH) extratropics (20°–90°S; Figure 1d). Examining the annual temperature trends, we find that the rate of global temperature increase has diminished when looking at more recent periods, and this phenomenon is seen outside of the tropics. Indeed, only the tropics display a significant warming trend for the 1999–2010 interval (Figure 1c) – elsewhere the trends are not significantly different from zero. The NH extratropics contains large trends (Figure 1b), as expected from examining land surface records, while the SH extratropical landmasses have much smaller trends (Figure 1d).
 Upon examining temperature trends seasonally, the global temperature record reveals that significant warming occurs in three seasons (boreal spring, summer, and fall) for the four earliest periods (i.e., 1979–2010, 1987–2010, 1993–2010, and 1999–2010). However, boreal winter (DJF) is the glaring exception in the record – only when starting from 1979–2010 is the DJF warming significant. For the following periods, the trend is no longer significant and even turns negative over the last decade. The only sub-region that mirrors this behavior in global temperature trends is the NH extratropics (Figure 1b).
 To check the robustness of this seasonal asymmetry and also examine potential influences of ocean temperatures on the global temperature trend, we repeat the analysis from Figure 1 but using NASA MERRA surface temperature data (Figure 2). Inclusion of sea surface temperatures (SSTs) with land temperatures yields similar trend results as from the land-only CRUTEM3 data – i.e., insignificant winter warming trends since the 1980s despite significant warming in the other seasons (Figure 2a). Global linear temperature trends are also dominated by the trends in the NH extratropics (Figures 2b and 2e). When examining changes in the trend in NH land and ocean temperatures, we find that SSTs exhibit robust, significant warming annually and seasonally as late as a start in 1993 (Figure 2e). Land regions show similar robust warming annually and for all seasons but winter, the only season with no statistically significant warming for any period (Figure 2b). The tropics have no significant trends in land temperatures or SSTs except when going back to 1979 (Figures 2c and 2f), and trends in the SH ocean and land are somewhat more random and even inconsistent between MERRA and the CRUTEM3 datasets (Figures 2d and 2g; differences are likely a function of different sampling). Staggering start and end dates by 1–2 years yields similar results for both NASA MERRA and CRUTEM3 (not shown). Hence, we confidently conclude that the cessation of global winter warming since 1987 is mostly attributable to neutral or even cooling temperature trends across the NH extratropical landmasses. We further computed the same trends using an observational ocean dataset [Smith et al., 2008] (shown in Figure S1 in the auxiliary material) and ERA-Interim [Dee et al., 2011] (not shown) and though trends differ slightly by region, the absence of winter warming is still tied to NH extratropical land cooling.
 Focusing on the NH extratropical land surface temperature trends since 1987, Figure 3shows both the area-averaged temperature anomaly seasonally (Figure 3a) and spatially (Figures 3b–3e). Rapid and significant warming is clearly evident in all seasons but winter, with the boreal fall land surface temperatures warming the most rapidly. For DJF, the linear temperature trend is nearly zero. Of importance is the spatial pattern of the linear temperature trends (Figures 3b–3e). For boreal summer and fall, except for select areas, most of the NH landmasses are experiencing strong warming trends, at or exceeding 1°C per decade in central and northern Eurasia (Figures 3d and 3e). Except for pronounced cooling in western and northwestern North America, boreal spring temperatures also are warming strongly (Figure 3c). However, corresponding to the near-neutral wintertime trend are large regions of significantly strong cooling trends across Europe, northern and central Asia, and parts of central and eastern North America (Figure 3b). This spatial pattern of temperature trends is reminiscent of the temperature regression pattern associated with the negative phase of the Arctic Oscillation (AO) [e.g., Thompson and Wallace, 2001]. Indeed, the pattern correlation between the winter temperature trend pattern and the temperature anomaly pattern associated with the negative phase of the AO is 0.85 (p < 0.01).
 The absence of a warming trend in winter is especially surprising given that coupled climate models project the strongest warming across the NH during boreal winter due to ‘winter (or Arctic, polar) amplification’ [Holland and Bitz, 2003; Alexeev et al., 2005; Langen and Alexeev, 2007; Serreze and Barry, 2011]. We computed the seasonal temperature trends and anomalies from the decadal1980 runs available from the CMIP5 model archive for the period 1987–2010. Independent of region or season, the models forecast positive temperature trends throughout all time periods, with insignificant warming only over the 2000s (see Figure S2). The model simulated interannual seasonal temperatures are compared with the NASA MERRA seasonal temperature anomalies in Figure 4. The annual ensemble-mean NH extratropical land surface temperatures (solid black line inFigure 4a) track well with the observed annual NH extratropical land surface temperatures (red line in Figure 4a), though the ensemble-mean slightly underestimates the total linear trend for the period (ΔTENSMEAN = 0.32°C versus ΔTOBS = 0.49°C). For a more complete comparison of the observed temperatures with the model simulated temperatures we include a spaghetti plot of all ensemble members in Figure S3.
 The models also simulate closely the seasonal trends for spring, summer and fall. For winter, while there is much more interannual variability in the observed versus simulated temperatures (the observed temperature falls outside the 1σenvelope two thirds of the time, double what is expected), the correspondence between DJF observed and simulated temperature trends remains the weakest of the four. Clearly, the difference in the boreal winter temperature trends represents some poorly resolved or missing forcing that is less influential in the other three seasons. We also note that the models predict robust cooling in all seasons due to the radiative forcing from the volcanic eruption of Mount Pinatubo. The simulated cooling is similar to the observations in all seasons except winter. This suggests that radiative forcing important in spring, summer and fall is likely masked by dynamic forcing associated with wave driving in winter. Ongoing research suggests that the models are deficient in simulating fall snow cover variability and wave forced stratosphere-troposphere coupling, which is contributing to poor model simulations of the AO and NH winter climate trends [e.g.,Hardiman et al., 2008; Cohen et al., 2012]. As the leading mode of NH wintertime climate variability, differences in the observed versus simulated AO are likely an important factor. Indeed, when examining the evolution of the AO and associated sea level pressure (SLP) trends across the NH, we find a divergence in the observed trends – the observations illustrate a trend toward negative AO conditions during DJF, while the models actually suggest a slight positive trend in the index (Figure S4a). Spatially, observations indicate statistically significant positive trends in SLP at high latitudes with statistically significant negative trends in SLP at lower latitudes (i.e., a negative AO pattern), while the models predict no significant SLP trends over the period.
4. Summary and Conclusions
 Analysis of monthly and annual temperatures over the past decade shows that the positive global temperature trend has become insignificant and small. Based on previously reported analysis of the observations and modelling studies this is neither inconsistent with a warming planet nor unexpected; and computation of global temperature changes over longer periods appropriate to the assessment of trends does exhibit statistically significant warming. However, upon examining the trends seasonally, more interesting and significant findings are discovered. In examining the NH extratropical landmasses, the biggest contributor to global temperature trends, we find substantial divergence in trends between boreal winter and the other three seasons. A statistically significant warming trend is absentacross NH landmasses during DJF going back to at least 1987, with either wintertime near-neutral or cooling trends. In contrast, significant warming is found for the other three seasons over the same time period.
 Based on current literature and our own examination of the latest coupled climate models, the lack of a significant warming trend in winter spanning nearly three decades is not likely or expected (less than 10% of the ensemble members analyzed in this study predicted no warming in winter). Therefore, we argue that any attribution study on the recent cessation of global warming should explicitly explain the seasonally asymmetric nature of the temperature trend. For example, studies that attribute the recent cooling to diminished shortwave radiation at the surface are at a great disadvantage since their influence is maximized during boreal summer and minimized during boreal winter, opposite to what has been observed.
 JLC is supported by the National Science Foundation grants ARC-0909459 and ARC-0909457 and NOAA grant NA10OAR4310163. MB is supported by National Science Foundation grant ARC-0909272. VA and JEC were supported by the National Science Foundation grants ARC 0909525 and Japan Agency for Marine-Earth Science and Technology. The authors acknowledge the climate modelling groups (listed inTable 1of this paper), the World Climate Research Programme's (WCRP) Working Group on Coupled Modelling (WGCM), and the Global Organization for Earth System Science Portals (GO-ESSP) for producing and making the CMIP5 model simulations available for analysis.
 The Editor thanks two anonymous reviewers for their assistance evaluating this paper.