Snow cover, with the largest areal coverage of any components in the cryosphere and high spatial and temporal variability, is believed to be an important indicator of climate change and has been shown to have strong impacts on regional and global energy and water cycles (Groisman et al., 1994; Qu and Hall, 2007; Hall et al., 2008). The heavy snow storms that have occurred in recent years led to significant economic loss and ecological impacts in southeastern Canada, where several observing stations broke long-standing snowfall records, resulting in significant impacts including roof collapses (Descurieux, 2010).
The annual maximum snow water equivalent (SWEmax) of snowpack is important for hydrological modelling and runoff prediction (Dery et al., 2005) and is also relevant to ground snow load calculations in the National Building Code of Canada (Newark et al., 1989). SWEmax represents accumulated snowfall events during snow season (defined as October to May in this paper) and is influenced by various metamorphosis processes associated with atmospheric variables (such as temperature, precipitation, and wind), land cover type and topography. Over North America, the main winter storm track is situated in an elongated area along the border between the United States and Canada, with a primary maximum centre situated over the Great Lakes and a secondary maximum centre over the lee side of the Rocky Mountains (Reitan, 1974), both areas being associated with preferable conditions of cyclogenesis.
Atmospheric teleconnection patterns characterize low-frequency climate variability on regional and hemispheric scales. The North Atlantic Oscillation (NAO) and the Pacific/North American pattern (PNA) are two of the most important patterns characterising the Northern Hemisphere extratropical climate variability, especially during the winter (Wallace and Gutzler, 1981; Jones et al., 1997), exerting significant influences on North American regions. Recent studies have revealed that the PNA and the NAO are two of the most important atmospheric low-frequency variability modes modulating snowpack interannual anomalies in North America via both snowfall and temperature pathways (Ghatak et al., 2010). Previous studies have shown that interannual variability in snowpack is highly correlated to local temperature and precipitation through controls of large-scale atmospheric circulations (Clark et al., 1999; Saito et al., 2004; Zhao and Fernandes, 2010). Understanding the role the PNA and the NAO play in the atmosphere-annual maximum snowpack interaction is important for improving the predictability of snowpack variability and its subsequent impact on infrastructure design under recent and future climate changes. However, only few studies (Jin et al., 2006; Brown, 2010) have been conducted on this linkage at regional scales.
The objective of this study is to investigate the relationship between the annual maximum snowpack (SWEmax) and the two atmospheric low-frequency variability modes, the NAO and PNA, using the Canadian Meteorological Centre (CMC) reanalysed SWE datasets covering snow seasons between 1979–1997 and 1998–2009. The study is focused on southern Canada, south of 55°N, where both the data quality and snow impacts are high. A potential change from pre-1998 to post 1998 in the SWEmax associated with the atmospheric circulation will be discussed via comparison analysis using annual maximum snow depth (SDmax) and the 850 hPa moisture transport field associated with the PNA/NAO modes.
2. Data and methods
Two gridded monthly SWE datasets used in this study are components of the historical daily snow depth analysis of Brown et al. (2003) covering North America for snow seasons 1979/1980–1996/1997 (P1) at a resolution of about 0.25° and the approximately 25 km CMC global operational daily snow depth analysis for snow seasons 1998/1999–2008/2009 (P2). SWE was estimated in P1 from snow aging expressions incorporated in the snowpack model used to derive the background field (Brown et al., 2003), while SWE in P2 was estimated using a monthly look-up table (Brown and Mote, 2008) developed for the snow-climate classes of Sturm et al. (1995). Over most of the Arctic region there are no observations, so these data are based almost entirely on estimated snow depths from the first guess field in that region (Brown et al., 2003; Brown and Mote, 2008). In this study, therefore, we limit our analyses to regions south of 55°N where data are constrained by a sufficient number of observing stations.
The concern is the possible impact of the discontinuity across 1998 resulting from the use of different snowpack models to generate the two SWE datasets. Therefore, an update version of Daily Gridded North American Snow Depth (Mote, 2008) at 1° resolution is used to verify the reality of the change in SWEmax observed from P1 to P2. The SWEmax and SDmax values are calculated during the snow season from October to May.
The NAO index is the normalized pressure difference between Gibraltar and SW Iceland (Jones et al., 1997), from the Climatic Research Unit, University of East Anglia. The PNA index is the second leading pattern obtained from a rotated principal component analysis of the 500 hPa geopotential height (GPH) field, from the NOAA Climate Prediction Center.
The monthly 850 hPa GPH, wind, specific humidity and water equivalent of accumulated snow depth (NARR SWE) fields from the North American Regional Reanalysis (NARR; Mesinger et al., 2006) at approximately 32 km resolution are used to produce winter (December to March) mean circulation fields and NARR SWEmax for the period 1979–2009. The 850 hPa water vapour transport field is simply calculated by multiplying respective horizontal wind components and specific humidity at each grid point. The time series of the indices and SWEmax/SDmax at each grid were de-trended before the analyses are carried out.
Figure 1 presents the mean SWEmax and the corresponding standard deviation (STD) for respective periods of P1 (Figure 1(a) and (c)) and P2 (Figure 1(b) and (d)). Some common spatial signatures shared by the two time periods are the high mean SWEmax values located over eastern Canada, spanning from southern Ontario to the Atlantic coast, and over the western cordillera. Similar geographical distribution signatures can be found in the interannual variability of SWEmax, with extreme values located close to the Pacific and Atlantic coastal regions and over the Rocky Mountains, showing evidence of sufficient water vapour sources and topographical lifting effects. The large variability over central Ontario, just north of the Great Lakes, and along St. Lawrence River, is connected to the maximum frequency of winter storm tracks. Comparing the two temporal periods, a reduction in the areal coverage of the high SWEmax and its interannual variability can be seen over eastern Canada from P1 to P2. A possible interpretation of this change is discussed below.
Figure 2 presents correlation coefficients of SWEmax associated with the reversed PNA and NAO indices for P1 and P2. Two correlation-sensitive regions can be identified, with one located over western Canada (Figure 2(a)), straddling the Rocky Mountains in central British Columbia and Alberta (and the Northwest United States), and the other one situated mainly over eastern Canada (Figure 2(c)), stretching from central and southern Ontario (north of the Great Lakes) to western Quebec. Both of these two regions are situated within the larger interannual variability regions discussed above (Figure 1). The statistically significant correlation regions associated with the winter PNA are more robust in the western domain while those with the winter NAO are more robust in the eastern domain; these two domains remain relatively sensitive to the two atmospheric low-frequency variability modes, compared to other regions. There are also areas of significant correlation over Northeast Quebec and north-central Canada (Figure 2(c)). For P2, the positive correlations detected during P1 with the PNA in the western domain have weakened (Figure 2(b)), while significant negative correlations with the NAO have emerged over northern Quebec and northwestern Ontario (Figure 2(d)).
To test whether or not the results from these two major correlation-sensitive domains are reflected in in situ observations, station-based time series of SWEmax from snow course data (Ross Brown, personal communication) for Yellowhead (52° 54′N, 118° 33 W) and Bigeast (45° 54′N, 79° 14′W) are presented with the reversed PNA and NAO indices in Figure 3(a) and (b). These in situ SWEmax time series are statistically significantly (98% confidence level) correlated to the PNA (−0.52) and the NAO (−0.41), respectively. Figure 3(c) shows the averaged SWEmax over central Ontario (the open square box in Figure 2(c) and 2(d)) on the base of the SWE datasets for P1 and P2 (blue lines) and the reversed NAO index (green) for joined P1 and P2. Statistical significance of this relationship is conducted via the Monte Carlo procedure by randomly shuffling the SWEmax time series 1000 times. The observed correlation against the distribution of correlations generated randomly between the SWEmax and NAO index is significant at p < 0.001 for P1 and p < 0.01 level for P2. Clearly, the statistically significant relationship between the SWEmax and the NAO index exists throughout the whole P1 + P2 period over this eastern NAO-sensitive spatial domain.
The concern is the possible impact of the discontinuity across 1998 resulting from the use of different snowpack models to generate the two SWE datasets. To address this issue, we have compared a reconstructed SWEmax time series (red lines in Figure 3(c), Ross Brown, personal communication), which used the same snowpack model that was employed to generate the first SWE dataset, with the two original SWEmax time series for P1 and P2. It is found that the reconstructed time series matches well with the two SWEmax time series for the overlapping period over the region of interest. To further test the robust results in Figure 2, we apply the same methodology to the SDmax, continuous time series covering 1979–2009 (Figure 4). The correlation patterns are very similar to those in Figure 2, indicating that the interannual variability in SWEmax is dominated by SDmax.
The NARR SWE data provide a continuous time series from 1979 to 2009, bridging the discontinuity in the CMC data across 1998. This allows us additional test for the validity in combining the two datasets. Correlation patterns using the NARR SWE data are shown in Figure 5. The overall patterns of correlation are similar to those shown in Figure 2, for before and after 1998, albeit with some regional differences. The sensitive regions in the east and west are still evident in Figure 5. Strong correlation with the NAO in the region north of the Great Lakes (Figure 5(c) and (d)) for the period before and after 1998 is consistent with what was shown in Figure 2. Similarly, the overall change in correlation (going from positive to negative across 1998 using the NARR SWE data) associated with the PNA is in general agreement with the result shown in Figure 2.
Figure 6 shows the averaged NARR SWEmax time series over central Ontario. The NARR data show a much lower SWEmax throughout the study period, compared to the SWEmax time series obtained from the CMC and a snowcourse data at Bigeast. However, the overall characteristics of the interannual variability of the three time series are similar across 1998. All the time series are in relatively good agreement for extreme events, such as those in 1982, 1997 and 2007.
To identify possible physical reasons for the results shown in Figures 2 and 3, we perform regression analyses of winter mean 850 hPa GPH and water vapour transport field versus the reversed winter PNA and NAO indices for each period. Figure 7(a) and (b) shows, for P1 and P2 respectively, a circulation anomaly pattern associated with the PNA, with three centres of action of opposite sign resembling the traditional PNA pattern. Between the northeastern Pacific centre and the centre over north-central Canada are the anomalous north-westerlies of water vapour transport providing water vapour sources for the Pacific coast and the lee side of the Rocky Mountains. The negative phase of the PNA with enhanced moisture transport is responsible for higher SWEmax over the western region, as is the case for P1. With the positive phase of the PNA, the associated anomalous water vapour transport is reversed with the southerly meridional transport that reduces winter precipitation (Jin et al., 2006). For P2 (Figure 7(b)), the continental centre over north-central Canada is weaker compared to that for P1, with the Pacific centre shifting westwards, leading to the weakening of water vapour transport toward the Pacific coast, resulting in reduced correlations with the snowpack over the western region of Canada. At the same time, the southeastern centre, moves northward resulting in an increased water vapour transport to Quebec; this is consistent with the occurrence of a strengthened correlation region in the western Quebec (Figure 2(b)).
Figure 7(c) and (d) shows a NAO-related circulation pattern characterized by a low centred over the Great Lakes, the region that showed significant correlation between the NAO and SWEmax during P1 (Figure 2(c)). During P2, the low circulation shifts to eastern United States with reduced strength and area. This is consistent with the smaller areal extent and the southward shift in the positive correlation regions over eastern Canada (Figure 2(d)). The regression pattern shown in Figure 7(d) for P2 is consistent with the negative NAO-SWEmax correlation pattern in the mid-to-high latitude regions (42°N and 55°N) in Figure 2(d). Over western Canada, the onshore water vapour transport anomaly during P1 is replaced by an offshore water vapour transport anomaly during P2, resulting in a change in correlation sign over the western domain. Over eastern Canada, the increased GPH does not facilitate precipitation and forces storm tracks to move southward, leading to negative correlations over the Hudson Bay coastal regions. The regime change in the atmospheric circulation associated with the NAO over North America is consistent with the change in the correlation pattern of the SWEmax over southern Canada associated with the NAO. This indicates that the change in the atmospheric circulations is a dominant factor influencing the correlation patterns of SWEmax with the atmospheric low-frequency variability modes over southern Canada.
4. Conclusion and discussion
In this study, we have explored the relationship between the SWEmax and the winter PNA and NAO using the SWE datasets covering 1979–1997 and 1998–2009. Two regions showing sensitivity in the correlation of SWEmax with the winter NAO and PNA indices have been identified, one centred over central Ontario and the other in western Canada (Rocky Mountains), with the latter one consistent with previous findings (Gutzler and Rosen, 1992) on the base of snow cover dataset. These regions are shown to have large SWEmax mean values and variability, corresponding to maximum frequency of winter storm occurrences. The correlations in the western domain are more robust when associated with the PNA, and those in the eastern domain are more robust when associated with the NAO. Similar results are found in SDmax and NARR SWE over the same interest period of 1979–2009, supporting the results in SWEmax.
In this study, the SWEmax over central Ontario is found to be significantly correlated with the NAO index throughout the interest period of 1980–2009, implying a nonstationary time series modulated by a covariate of the NAO index for the generalised extreme value analysis (Coles, 2001). The high stability may be attributed to the geographically locked maximum frequency of winter storm occurrences and effects of the Great Lakes (Burnett et al., 2003). Using a reconstructed SWEmax time series for 1972–2002 over this region, a NAO index-based linear regression model is found to explain 47% of the variance of the time series while a combined NAO–PNA regression model is found to explain an additional variance of less than 10%. Brown (2010) using a reconstructed SWE dataset over Quebec for 1949–2005 claimed that lack of robust correlations between the SWEmax and the NAO for 1950–2005 can be attributed to the decadal and interdecadal changes in the low-frequency variability modes. Our results over Quebec are consistent with his study (Brown, 2010) in that the positive phase of the NAO is associated with a northwesterly flow over northern Quebec with cold and dry air, resulting in low SWEmax, and a statistically significant negative correlation exists only prior to 1997 over northwestern Quebec.
The change with weakened correlation in the western domain and southward shift of significant correlation extent in the eastern domain, associated with the respective PNA and NAO for the post-1998 period, has been found to be associated with an atmospheric circulation change, as indicated by the results of analyses of the NARR 850 hPa GPH and water vapour transport fields. The weakened correlation of SWEmax with the PNA in the western domain during the post-1998 period is associated with a weakening of the PNA centre of action over north-central Canada and a westward shifting of the Northeast Pacific centre of action, leading to reduced moisture over the Pacific coast and the lee side of the Rocky Mountains. The southward shift in the reduced correlation extent in the eastern domain with the NAO during the post-1998 period is associated with a southward shift in the weakening continental NAO centre over the Great Lakes. Furthermore, the occurrence of the strengthened opposite correlations over northern Canada (above 53°N) associated with the NAO during this post-1998 period may also be attributed to the atmospheric circulation change, which forms a more zonal circulation pattern in the lower troposphere, resulting in anomalous water vapour transport over the west coast and southern Hudson Bay. Zhang et al. (2008) reported that the traditional tri-polar Arctic Oscillation/NAO was transformed into a dipolar structure between the Eurasian Arctic coast and North Pacific in the early 2000s. The pronounced change in the atmospheric circulation patterns which occurred after 1998 may be an integrated part of the recent climate change associated with the observed long-term warming Arctic and declining sea-ice extent (Zhang et al., 2008), which perhaps has influenced the extreme snowpack distribution.
Given the results of a close connection between the phases of the NAO, with the resulting corresponding circulation anomalies, and the occurrences of SWEmax over certain regions in Canada, it is of interest to employ this connection to speculate on the future change in the frequency of SWEmax. In this context, we refer to the study conducted by Wang et al. (2010) who found that during a ‘longer-term lower’ NAO phase, extreme negative NAO events are more likely to occur than extreme positive NAO events. Since 1998, the interdecadal NAO variability has been in the negative phase, and if this continues there will be more frequent occurrences of extreme negative NAO events, which will result in an increased frequency of high SWEmax events and may enhance snow load risk to infrastructure.
The authors would like to thank Ross Brown for generously providing the SWE data and important suggestions. Thanks are also due to Drs. Ken Yuen, and Shunli Zhang for their insightful comments and data preprocessing. This study is supported by Environment Canada snow load project.