Origin and predictability of the extreme negative NAO winter of 2009/10



[1] The winter of 2009/2010 was one of the most negative winters of the North Atlantic Oscillation (NAO) during the last 150 years. While most operational extended-range forecasting systems had difficulties in predicting the onset of the negative NAO phase, once established, extended-range forecasts were relatively skilful in predicting its persistence. Here, the origin and predictability of the unusual winter of 2009/10 are explored through numerical experimentation with the ECMWF Monthly forecasting system. More specifically, the role of anomalies in sea surface temperature (SST) and sea ice, the tropical atmospheric circulation, the stratospheric polar vortex, solar insolation and near surface temperature (proxy for snow cover) are examined. None of these anomalies is capable of producing the observed NAO anomaly, especially in terms of its magnitude. The results of this study support the hypothesis that internal atmospheric dynamical processes were responsible for the onset and persistence of the negative NAO phase during the 2009/10 winter.

1. Introduction

[2] The winter season of 2009/10 was one of the coldest and snowiest in central North America and northwestern Europe for decades and it was suggested that this was a result of the extremely negative phase of the NAO [Seager et al., 2010]. The so-called ECMWF climagrams show that the observed NAO index lay below the 5-th percentile for all of the three winter months December 2009 to February 2010, with February being the most extreme of the three months (Figure 1a). Interestingly, the NAO remained negative throughout the subsequent spring season, although with reduced amplitude.

Figure 1.

ECMWF climagram showing the distribution of monthly mean values of the NAO index for operational ECMWF seasonal forecasts (purple boxes and whiskers) started on (a) 1 November 2009, (b) 1 December 2009 and (c) 1 January 2010, the corresponding model climatology derived from hindcasts (1981–2005, light-blue boxes and whiskers), and observations for a 25-year period (yellow and orange band). The limits of the purple/grey whiskers and yellow band correspond to the 5-th and 95-th percentiles, those of the purple/grey box and orange band to the lower and upper tercile, while the median is represented by the line within the purple/blue box and orange band. Observed monthly mean values for the winter of 2009/10 are also shown (red circles). The NAO index is computed by projecting both observed and predicted monthly mean Z500 anomalies computed relative to the period 1981–2005 onto the first EOF of observed wintertime North Atlantic Z500 anomalies from the period 1981–2005.

[3] In terms of extended-range forecast skill the winter of 2009/10 can be divided into two parts. During late autumn and early winter the extended-range forecast skill of the ECMWF seasonal forecasting system in predicting the negative NAO winter was poor to modest (Figures 1a and 1b). However, once the negative phase of the NAO was well established the extended-range forecast skill increased substantially (Figure 1c and forecasts started on 1 February 2010, not shown).

[4] What caused the unusually extreme and persistent negative NAO phase during the winter of 2009/10? A reasonable null hypothesis is to assume that the negative NAO phase was a result of atmospheric non-linear dynamical processes which are known to explain a large amount of interannual atmospheric variability in the extratropics [e.g., James and James, 1989], including that of the NAO [Kushnir et al., 2006]. However, given the outstanding magnitude and persistence of the negative NAO winter in 2009/10 it is sometimes argued that some ‘external’ forcing must have contributed. Here ‘external’ shall refer to anything beyond tropospheric dynamics in the North Atlantic region.

[5] A number of ‘external’ phenomena, which prevailed during the 2009/10 winter, have been suggested in the past to increase the frequency of occurrence of the negative NAO phase. The list of phenomena considered in this study comprises El Niño [Fraedrich, 1994; Greatbatch and Jung, 2007; Brönnimann, 2007; Ineson and Scaife, 2009], the negative phase of the Quasi-Biennial Oscillation (QBO) [e.g., Boer and Hamilton, 2008], a weakened stratospheric polar vortex [Baldwin and Dunkerton, 2001], reduced incoming solar radiation (for a review see Gray et al. [2010]) and Northern Hemisphere snow cover anomalies [Cohen et al., 2010].

[6] The aim of this study is to explore possible causes and the predictability of the negative NAO winter 2009/10.

2. Methods

[7] Seasonal forecasting experiments presented in this study were carried out using the ECMWF Monthly Forecasting system (MOFC) [Vitart, 2004], which is based on a coupled atmosphere-ocean model. The atmospheric component consists of a recent version (36r1) of the ECMWF Integrated Forecasting System at TL159 with 60 levels in the vertical. The ocean component consists of the Hamburg Ocean Primitive Equation (HOPE) model at 1° × 1° horizontal resolution and with 29 levels in the vertical. Ensemble forecasts (40 members) for the winter of 2009/10 (December through February) were started on 1 November 2009. In order to be able to compute anomalies, for each of the different experiment types, 4 member ‘calibration’ ensembles were also produced for all winters of the period 1991–2008. Details about how the ensembles were generated are described by Vitart [2004].

[8] In order to explore whether the negative phase of the NAO was caused by atmospheric circulation anomalies in certain ‘remote’ regions such as the tropics, a relaxation approach was employed; that is, during the integration the model was relaxed towards interpolated, 6-hourly ECMWF analysis fields of zonal and meridional wind, temperature and surface pressure in certain regions using a relaxation time scale of 10 hrs (see Jung et al. [2010a, 2010b] for details).

3. Results

3.1. Before the Onset

[9] Observed 500 hPa geopotential height (Z500) anomalies for the winter of 2009/10 are shown in Figure 2a. The negative phase of the Pacific-North America (PNA) pattern and the NAO are clearly evident. Corresponding Z500 anomalies obtained from seasonal forecasts with the coupled atmosphere-ocean version of the ECMWF MOFC are shown in Figure 2b. The coupled model shows some skill in simulating the PNA response to the El Niño conditions in the tropical Pacific. It fails, however, in predicting the negative phase of NAO.

Figure 2.

Ensemble mean geopotential height anomalies (in m) at 500 hPa for the winter (December through February) 2009/10: (a) observed, (b) coupled control integration, (c) uncoupled control integration with observed SST and sea ice, (d) tropical relaxation, (e) relaxation in the East Asia-Western North Pacific region, relaxation of the Northern Hemisphere stratosphere for model levels (f) 1–17 (about 0.1–30 hPa) and (g) 1–22 (about 0.1–85h Pa), (h) relaxation of near-surface temperatures (five lowest model levels from about 990–1013 hPa) over the Northern Hemisphere and (i) experiment with reduced incoming UV radiation. Notice the larger contour interval in Figure 2a. Differences statistically significant at the 95% confidence level in Figures 2b–2i are hatched. Anomalies in Figures 2b–2i were computed as the difference between the ensemble mean for 2009/10 (40 members) and the long-term mean from respective 4 member hindcast integrations covering the period 1991/92–2008/09 (one separate hindcast for each experiment type).

[10] In order to determine whether the lack of seasonal predictive skill is due to the failure of the coupled model to predict the observed SST and sea ice anomalies, an additional seasonal forecast experiment has been carried out in which the uncoupled atmosphere-only version of the ECMWF model has been forced by observed daily SST and sea ice fields (Figure 2c). The uncoupled integration produces small (but significant) Z500 anomalies, which seem to be in quadrature with the observed anomalies. These results suggest that neither SST nor sea ice anomalies explain the negative phase of the NAO during the 2009/10 winter.

[11] It could be argued, of course, that the atmospheric component of the ECMWF model fails to realistically respond to the observed SST and/or sea ice conditions. In order to test this conjecture for the tropics, where the ocean-atmosphere forcing on seasonal time scales is believed to be at its strongest, additional experiments with the coupled ECMWF model were carried out in which the tropical atmosphere (20°S–20°N) was relaxed towards 6-hourly analysis fields (see Jung et al. [2010a, 2010b] for details). Prescribing the observed evolution of the tropical atmosphere during the course of the seasonal forecasts improves the atmospheric response over the North Pacific; over the North Atlantic, however, the simulated atmospheric response shows little resemblance with what was observed during the winter of 2009/10, at least in terms of its magnitude (Figure 2d). Notice, that tropical relaxation also imposes the negative phase of the QBO as well as the Madden-and-Julian Oscillation both which are believed to exert some influence onto the NAO [Boer and Hamilton, 2008; Cassou, 2008]. Interestingly, similar experiments carried out to explain the negative NAO winter 2005/06 led to Z500 anomalies similar in magnitude to those observed [Jung et al., 2010b]. This suggests that the relaxation approach is able to distinguish different origins of NAO anomalies.

[12] It is possible that the extratropical response to a tropical forcing is not realistically captured by the ECMWF model. In order to ensure that such a possible model problem does not affect our conclusions an additional relaxation experiment has been carried out in which the atmosphere was relaxed to analysis data in the East Asia-Western North Pacific (EAWNP, 20°–60°N and 80°–150°E) region (Figure 2e), that is, in a region well known for the occurrence of strong tropical-extratropical interactions. This additional experiment suggests that ‘perfect knowledge’ of the anomalous flow conditions in the southeast Asian wave guide and the entrance region of the North Pacific storm track, whether tropically forced or not, is not sufficient to reproduce the negative phase of the NAO during the winter of 2009/10.

[13] Given previous research [e.g., Baldwin and Dunkerton, 2001; Douville, 2009], it could be argued that the anomalously weak stratospheric polar vortex during the winter 2009/10 contributed to the negative phase of the NAO. In order to test this possibility, two additional seasonal forecasting experiments with the coupled ECMWF model have been carried out in which the Northern Hemisphere (30°–90°N) stratosphere was relaxed towards analysis data upward of about 30 hPa (Figure 2f) and upward of about 85 hPa (Figure 2g). Like for the other experiments, these stratospheric relaxation experiments fail to reproduce the magnitude of the observed NAO anomaly (Figure 2f). Both experiments employ a smooth transition zone in the vertical for the relaxation [Jung et al., 2010a]. For the experiment with relaxation of model levels 1–22 this means that the relaxation coefficient decreases from 0.05 hr−1 at about 55 hPa to 0.01 hr−1 at about 85 hPa. It is conceivable that stronger relaxation of the lower stratosphere would give a larger ‘signal ’in the troposphere [Douville, 2009]. However, this would also pose the risk of explicitly forcing the upper troposphere. We have repeated the above experiments by relaxing only stratospheric zonal winds, which are better analysed by data assimilation systems. Relaxing zonal winds only, however, does not change the conclusions of this study (not shown).

[14] It has been argued that the negative NAO winter 2009/10 can be explained by Eurasian snow cover anomalies [Cohen et al., 2010]. To test this hypothesis, another seasonal forecast experiment has been carried out in which temperature in the lowest five model levels (about lowest 300 m) has been relaxed towards analysis data over the Northern Hemisphere in order to prescribe (amongst others) snow-driven temperature anomalies on the lower part of the planetary boundary layer. The resulting Z500 anomalies (Figure 2h) show little resemblance with the observations. The spatial structure of the Z500 response, for example, is in quadrature with the observed Z500 anomalies. The implied weak role of snow cover anomalies is consistent with other research: Firstly, additional seasonal forecast experiments with the ECMWF model for the winter of 2009/10 with different snow initial conditions indicate that the impact on early winter snow was negligible (G. Balsamo, personal communication, 2011). Secondly, snow appears to affect the atmosphere primarily in late spring, summer and early autumn rather than winter [e.g., Bojariu and Gimeno, 2003].

[15] It is increasingly being accepted that solar variability has an influence on climate including the NAO [e.g., Gray et al., 2010; Lockwood et al., 2010; Woollings et al., 2010]. Given the existence of such a link together with the fact that solar activity in 2009/10 had fallen to values unknown since the start of the 20th century [Lockwood et al., 2010], it seems natural to explain the negative NAO winter 2009/10 by unusually low solar activity. In order to test this conjecture, additional experiments were carried out in which the incoming solar radiation in the UV-B and UV-C band has been reduced by 4% and 6%, respectively. This reflects the usual drop in UV observed over the last two 11-yr cycles. Changing solar radiation in the UV part of the spectrum only rather than the solar constant reflects the fact that most of the solar variability associated with the 11-yr cycle is concentrated in the UV band [Gray et al., 2010]. Notice, that all integrations were carried out with interactive radiation and prognostic ozone. The experiments carried out in this study suggest that the impact of anomalously low incoming UV radiation on the tropospheric circulation in the North Atlantic region are very small (Figure 2i) suggesting that the unusually low solar activity contributed little, if any, to the observed NAO anomaly during the 2009/10 winter.

3.2. After the Onset

[16] Operational monthly forecasts carried out with the ECMWF Variable Resolution Ensemble Prediction System (VAREPS) [Vitart et al., 2008] in January and February 2010 were very skilful in predicting the negative phase of the NAO three to four weeks in advance (Figures 3a, 3b, 3d, and 3e). This large skill appears to be primarily associated with the unusually persistent character of the NAO during this period. The high skill of these forecasts suggests that the ECMWF VAREPS, which is similar to the model used in the previous section except for higher horizontal resolution, realistically captures the relevant processes contributing to the atmospheric persistence in the North Atlantic region in mid to late winter 2009/10. This fact has been exploited to further test the role of SST and sea ice anomalies in explaining the origin of the persistence of the negative NAO phase. For each of the two forecasts considered here (forecasts started on 14 January and 11 February 2010, respectively) the initial conditions for ocean temperatures and sea ice were replaced by those taken from 1990, that is, from one of the strongest positive NAO winters during the last 150 years. The fact that the forecasts in the North Atlantic-European region are virtually unchanged in these experiments (Figures 3c and 3f) suggests that SST and sea ice anomalies, especially those in the North Atlantic region, were not crucial for explaining the persistence of the negative phase of the NAO once it was established. Similar results were obtained with ocean and sea ice initial conditions taken from January and February of the years 1994, 2005 and 2006 (not shown).

Figure 3.

Geopotential height anomalies at 500 hPa (in m). (a) Observed anomalies for the period 1–15 February 2010 (verifying analysis); corresponding D+18–D+32 forecasts started on 14 January 2010: (b) control, (c) ocean initialization using data from 1990. (d) Observed anomalies for the period 1–15 March 2010; corresponding D+18–D+32 forecasts started on 11 February 2010: (e) control and (f) ocean and sea ice initialization using data from 1990. Anomalies were computed with respect to the period 1992–2009.

4. Discussion

[17] The origin and predictability of the extremely negative phase of the NAO during the winter of 2009/10 have been explored by means of numerical experimentation. Different possible forcing mechanisms have been tested such as El Niño, the QBO, and reduced solar insolation. However, none of the forcings considered in this study was able to reproduce the negative phase of the NAO, especially in terms of its magnitude. The results of this study, therefore, support the hypothesis that both the development and persistence of negative NAO phase resulted from internal atmospheric dynamical processes. This may explain why most operational seasonal forecasting systems had problems in predicting the negative NAO winter when started in late autumn (e.g., forecasts issued in October and November 2009). The results of this study suggest that internal atmospheric dynamics are an important source of low-frequency atmospheric interannual variability [see also James and James, 1989] including extreme atmospheric circulation anomalies. Furthermore, this study suggests that internal atmospheric dynamics are able to produce extremely persistent atmospheric circulation anomalies which are associated with substantial extended-range predictive skill. From a dynamical point of view it will be interesting to better understand the processes responsible for the persistence of the negative NAO throughout most of the 2009/10 winter.


[18] We acknowledge access to operational seasonal forecast products from Meteo-France and the Met Office through EUROSIP. The authors benefitted from useful suggestions by Adam Scaife and two anonymous reviewers.

[19] The Editor thanks the two anonymous reviewers for their assistance in evaluating this paper.