Abstract

This study assessed the climate and trend of cyclone activity in Canada using mainly the occurrence frequency of cyclone deepening events and deepening rates, which were derived from hourly mean sea level pressure data observed at 83 Canadian stations for up to 50 years (1953–2002). Trends in the frequency of cyclone activity were estimated by logistic regression analysis, and trends of seasonal extreme cyclone intensity, by linear regression analysis.

The results of trend analysis show that, among the four seasons, winter cyclone activity has shown the most significant trends. It has become significantly more frequent, more durable, and stronger in the lower Canadian Arctic, but less frequent and weaker in the south, especially along the southeast and southwest coasts. Winter cyclone deepening rates have increased in the zone around 60°N but decreased in the Great Lakes area and southern Prairies–British Columbia. However, extreme winter cyclone activity seems to have experienced a weaker increase in northwest-central Canada but a stronger decline in the Great Lakes area and in southern Prairies. The results also show more frequent summer cyclone activity with slower deepening rates on the east coast, as well as less frequent cyclone activity with faster deepening rates in the Great Lakes area in autumn.

Cyclone activity in Canada was found to be closely related to the North Atlantic Oscillation (NAO), the Pacific Decadal Oscillation (PDO), and El Niño–Southern Oscillation (ENSO). Overall, cyclone activity in Canada is most closely related to the NAO. The simultaneous NAO index explains about 44% (41%) of the winter (autumn) cyclone activity variance in the east coast, 31% of winter cyclone activity variance in the 60°–70°N zone, and 17% of autumn cyclone activity variance in the Great Lakes area. Also, in several regions (e.g., the east coast, the southwest, and the 60°–70°N zone) up to 15% of the seasonal cyclone activity variance can be explained by the NAO/PDO/ENSO index one–three seasons earlier, which is useful for seasonal forecasting.

1. Introduction

Cyclone activity is usually accompanied with adverse weather conditions and, hence, plays an important role in the climate system. A systematic change in either the geographical location or the intensity of cyclone activity will result in substantial precipitation anomalies, among other impacts on regional climates. Assessment of changes in severe weather phenomena such as those associated with strong cyclone activity is important to socioecosystems.

It is well known that cyclones may vary in intensity, duration, frequency, and trajectory among other characteristics. Due to the complexity of cyclones, a number of cyclone activity indices have been developed, mainly including (but not limited to) cyclone count statistics and eddy variance/covariance statistics. However, a given index may be closely related to some, but not all, aspects of cyclone activity. Certain indices may better reflect the impacts of cyclones on human society and ecosystems whereas others are better suited to understand dynamics (Paciorek et al. 2002). Which aspects of cyclone activity to focus on, or what indices to use, also depends on the type of data available for analysis. The range of fields used for characterizing cyclone activity include 1) mean sea level pressure (MSLP); 2) different levels of tropospheric height, meridional wind, vorticity, and temperature; and 3) potential vorticity and potential temperature on a selected surface (e.g., Raible and Blender 2004; Leckebusch and Ulbrich 2004; Hodges et al. 2003; Hoskins and Hodges 2002; Sickmoller et al. 2000; Schubert et al. 1998; Blender et al. 1997). Among these fields, MSLP is probably the most popularly used data for studying cyclone activity (e.g., Wang et al. 2006, 2004; Lambert 2004; Zhang et al. 2004; Fyfe 2003; Zolina and Gulev 2002; Graham and Diaz 2001; Gulev et al. 2001; Simmonds and Keay 2000; Rogers 1997; Serreze et al. 1997, 1993; Lambert 1996; Serreze 1995; among others). This study is also based on MSLP data, as described in section 2 below.

There have been many studies using global reanalysis data to assess observed changes in boreal extratropical cyclone activity (e.g., Wang et al. 2006, 2004; Zhang et al. 2004; Hodges et al. 2003; Paciorek et al. 2002; Graham and Diaz 2001; Gulev et al. 2001; Sickmoller et al. 2000; Blender et al. 1997; Serreze et al. 1997, 1993; Lambert 1996; Serreze 1995). In particular, a recent study by Hodges et al. (2003) compiled a comprehensive comparison between cyclone tracking statistics of four reanalysis datasets for the period 1979–96. Most recently, Wang et al. (2006) compared the climatology and changes of extratropical storm tracks and cyclone activity derived from the 6-hourly MSLP fields of the European Centre for Medium-Range Weather Forecasts (ECMWF) 40-yr Re-Analysis (ERA-40) (Uppala 2001) with those derived from the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis (NNR; Kalnay et al. 1996) for the period 1958–2001. They reported that both ERA-40 and NNR show significant increases in the number of winter cyclones in the lower Canadian Arctic, with significant reductions in southern Canada (cf. Fig. 1). Figure 2a shows the time series of log odds of winter cyclone occurrence in the Canadian 60°–70°N zone (140°–65°W) as derived from the ERA-40 and NNR MSLP data (i.e., the cyclone database of Wang et al. 2006), which has a highly significant upward trend [with significance (1 − p) = 0.009 estimated from the ERA-40 data] and is also highly correlated with the simultaneous North Atlantic Oscillation (NAO) index (shown in Fig. 2b). These changes are consistent with those reported in previous studies. For example, Gulev et al. (2001) analyzed the NNR MSLP data and reported an increase in the number of winter cyclones over the same lower Canadian Arctic region, with a decreasing tendency in southern Canada (especially the Canadian east coast; cf. Figs. 5c, d in Gulev et al. 2001). A significant increase in Arctic cyclone activity during the second half of the twentieth century, with a decrease in the midlatitude cyclone activity, was also reported by Zhang et al. (2004) and Serreze et al. (1997) through analyzing the NNR MSLP data or the National Meteorological Center (NMC) twice-daily MSLP data.

Fig. 1.

Changes in the 20-yr total counts of winter (JFM) cyclones over Canada from 1958–77 to 1982–2001 (1982–2001 minus 1958–77), as derived from (a) ERA-40 and (b) NNR data (i.e., the cyclone database of Wang et al. 2006). Hatching indicates areas where changes are of at least 5% (gridded) and 20% significance.

Fig. 1.

Changes in the 20-yr total counts of winter (JFM) cyclones over Canada from 1958–77 to 1982–2001 (1982–2001 minus 1958–77), as derived from (a) ERA-40 and (b) NNR data (i.e., the cyclone database of Wang et al. 2006). Hatching indicates areas where changes are of at least 5% (gridded) and 20% significance.

Fig. 2.

Time series of (a) the areal mean log-odds of winter cyclone occurrence in the Canadian 60°–70°N zone from 140° to 65°W) as derived from the ERA-40 or NNR MSLP data (i.e., the cyclone database of Wang et al. 2006; the trend line is estimated from ERA-40; but it is similar for NNR), (b) winter (JFM) mean NAO index, and (c) the areal mean log-odds of winter cyclone deepening events in the Canadian 60°–70°N zone. Correlation coefficients between the log-odds time series and the NAO index, as well as their p values, are also given in parentheses.

Fig. 2.

Time series of (a) the areal mean log-odds of winter cyclone occurrence in the Canadian 60°–70°N zone from 140° to 65°W) as derived from the ERA-40 or NNR MSLP data (i.e., the cyclone database of Wang et al. 2006; the trend line is estimated from ERA-40; but it is similar for NNR), (b) winter (JFM) mean NAO index, and (c) the areal mean log-odds of winter cyclone deepening events in the Canadian 60°–70°N zone. Correlation coefficients between the log-odds time series and the NAO index, as well as their p values, are also given in parentheses.

In the meantime, there has arisen the question of “whether or not the trends are just an artifact of the reanalysis procedure or the increasing availability of data (including satellite data) in recent decades.” Whenever and wherever possible, one should check the veracity of trends estimated from reanalysis data. This can be done by comparing trends estimated from reanalysis data with those estimated from in situ observations when the relevant in situ data are available. For example, the changes in winter cyclone activity in Canada derived from the NNR or ERA-40 data (Wang et al. 2006; Gulev et al. 2001) can be verified using directly the in situ MSLP data from Canadian stations that have been homogenized to the extent possible (although most of the raw SLP data might have been assimilated in the two reanalyses). However, the Canadian in situ observations of surface atmospheric pressure data has never been used directly to assess observed changes in cyclone activity in Canada. Actually, a trend assessment devoted specifically to cyclone activity in Canada has not been done before, although some previous studies included Canada as part of the boreal extratropics (e.g., Wang et al. 2006; Gulev et al. 2001; Lambert 1996) or the Arctic (e.g., Zhang et al. 2004; Serreze et al. 1997, 1993; Serreze 1995).

The first objective of the present study is to check the veracity of changes in cyclone activity in Canada that are seen in the global reanalyses (NNR or ERA-40) data. Hourly observations of surface atmospheric pressure from 83 Canadian weather observing sites for the period 1953–2002 are analyzed to assess the observed changes, which are then compared with the changes derived from the NNR and ERA-40 data. The major aspects of the assessment include changes in the seasonal mean and variance of cyclone deepening rates and in the frequency, duration, and intensity of cyclone activity.

Storm track activity over the Northern Hemisphere is connected to the dominant patterns of atmospheric variability, such as the NAO. Positive anomalies of the NAO index (Hurrell 1995) are associated with the strengthening of the midlatitude westerly flow over the North Atlantic, and should lead to the intensification and poleward deflection of the North Atlantic midlatitude storm track (Gulev et al. 2001). The NAO index was found to be positively correlated with the frequency of winter cyclones in Canada (cf. Figs. 2b,c) and also with the frequency of deep cyclones (deeper than 980 hPa) over North America (mainly southern Canada) and the European Arctic (Gulev et al. 2001). By analyzing the 1000-hPa geopotential heights taken from the ECMWF reanalyses for 1979–97, Sickmoller et al. (2000) also reported relationships between cyclone activity in the Northern Hemisphere to the NAO and the El Niño–Southern Oscillation (ENSO). Using the NMC MSLP fields for 1966–93, Serreze et al. (1997) reported that, during negative extremes of the NAO, there is a twofold decrease in cold season (October–March) cyclone events within the climatological Icelandic low region and a modest increase in cyclone activity to the south over a large area from Labrador southeastward to Portugal.

The impacts of NAO, ENSO, and the Pacific Decadal Oscillation (PDO) on the climate and its variability in Canada are well known. For example, Higuchi et al. (2000) found that three of the four past major freezing rain events in south-central Canada since 1958 were associated with the positive phase of the NAO. It has also been diagnosed that the ENSO plays a dominant role in Canadian temperature and precipitation anomalies, especially in winter (Shabbar and Barnston 1996; Shabbar et al. 1997). Huang et al. (1998) showed significant coherence between NAO and Niño-3 sea surface temperature (SST) in about 70% of the warm ENSO events from 1900 to 1995. Results of Bonsal et al. (2001) show the NAO as the dominant low-frequency variability mode affecting winter temperature, with the effects mainly confined to northeast Canada. Bonsal et al. reported that the ENSO and PDO influences are somewhat weaker and occur over western and central Canada and that the associated winter PDO pattern has a significant modulating effect on ENSO-related temperature responses. All these observed relationships are explained by variations in the associated midtropospheric circulation over the North Pacific and western and central North America (Bonsal et al. 2001).

In this context, it is important to characterize the relationships between cyclone activity in Canada and major circulation regimes, which would be also helpful for understanding the observed changes, and for cyclone activity forecast and weather-related risk management. Thus, in addition to the assessment of changes observed in the last half century, the present study also aims to explore the relationships between cyclone activity in Canada and a few major circulation regimes that are known to be well associated with weather and climate in Canada (see section 5 below for more details).

The remainder of this paper is arranged as follows. The data, as well as the trend analysis and data homogenization methods used in this study, are described in sections 2 and 3, respectively. The climatology and observed changes of cyclone activity are presented and compared with those derived from reanalysis data in section 4. The relationships between the frequency of cyclone activity and a few major circulations regimes are explored in section 5. This study is completed with a summary and some discussion in section 6.

2. Data

The database for this study consists of surface atmospheric pressure observed in Canada during the period 1953–2002. Surface atmospheric pressure is usually recorded for both station and mean sea level, which are called station pressure and MSLP, respectively. Station elevation is critical in the calculation of station pressure and MSLP values from barometer readings. Changes in the “elevation definition” and in the algorithms used to calculate the station pressure correction and the station pressure to MSLP reductions, such as those took place in November 1976 and January 1977 in Canada, could cause discontinuities (mean shifts or step changes) in both station pressure and MSLP time series and hence could affect estimates of trends, especially in southern Canada (Slonosky and Graham 2005).

Therefore, in this study, we used 3-hourly MSLP changes (also called MSLP tendencies) calculated from hourly MSLP data observed at 83 weather observing stations in Canada. Our choice for using MSLP changes was based on the consideration that strong cyclones or severe storms are usually associated with large surface atmospheric pressure changes within a few hours and that any data discontinuity problem can be greatly diminished by using pressure changes (because a mean shift in the pressure time series will affect only one single pressure change value, i.e., one difference between two measurements 3-h apart, which is usually easy to detect and has little effect on trend estimate). Also, for the study of cyclone activity, it is more convenient to use pressure at one specific level, such as mean sea level. Thus, MSLP changes were preferred over station pressure changes here. Nevertheless, before the calculation of 3-hourly MSLP changes, we visually compared the MSLP time series with the relevant station pressure series to eliminate possible “transposition” between them (Slonosky and Graham 2005) and to remove obvious discontinuities in the MSLP time series, although the hourly MSLP data from the National Archives of the Meteorological Service of Canada had undergone a quality control procedure. Further, we visually inspected time series of the hourly MSLP and 3-hourly MSLP changes to remove obvious outliers therein and to check for a second time for any data discontinuity. When necessary, time series of other variables such as precipitation and temperature observed at the same site were also checked to help determine if a value is an outlier or a real extreme value. In other words, both the MSLP data and 3-hourly MSLP changes used in this study are subject to a preliminary homogenization procedure prior to the calculation of any cyclone activity index.

To use the maximum sample size possible, we calculated 3-hourly MSLP changes for all moving 3-h windows (i.e., hours 0000–0300, 0100–0400, . . . 2100–0000, 0000–0300, and so on). Thus, there should be (n − 3) samples for a station of n hours continuous observations (without any missing; n = Nd × 24 h for a station of complete Nd days of observations).

As shown in Fig. 3, among the 83 stations, 78 of them have nearly complete, continuous 3-hourly MSLP change records for the 50-yr period from 1953 to 2002, the other 5 stations also have nearly complete, continuous records but for shorter periods ranging from 30 to 49 years (here “nearly complete” means with less than 10% missing values). These stations of shorter records of MSLP data were included in this study to provide a better spatial coverage (mainly to fill in large data gaps in northern Canada; cf. Fig. 3).

Fig. 3.

Locations of the 83 stations of long-term hourly MSLP data used in this study, and the four selected areas: east coast (south of 52°N and east of 69°W), Great Lakes (south of 52°N and 69°–85°W), southwest (south of 60°N and west of 85°W), and the 60°–70°N zone. Circles indicate the five stations of shorter (30–49 yr) MSLP data series.

Fig. 3.

Locations of the 83 stations of long-term hourly MSLP data used in this study, and the four selected areas: east coast (south of 52°N and east of 69°W), Great Lakes (south of 52°N and 69°–85°W), southwest (south of 60°N and west of 85°W), and the 60°–70°N zone. Circles indicate the five stations of shorter (30–49 yr) MSLP data series.

In this study, cyclone activity is said to be present at a site when the MSLP at that site is below 1000 hPa. Although cyclones are not really tied to any particular pressure value, we only focus on these cyclones in this study (we also tried with a set of different pressure threshold values and found that the results are not very sensitive to the threshold value selected, as long as the selected threshold value is high enough to include the vast majority of cyclones). In the presence of cyclone activity, the magnitude of 3-hourly MSLP changes reflects the intensity of cyclone activity, and the number of MSLP changes exceeding an extreme magnitude reflects the frequency of strong cyclone activity. Here, we analyzed only those MSLP changes that were associated with cyclone activity, that is, only those that took place in the presence of a MSLP below 1000 hPa at the site. That is, we defined cyclone deepening rates as

 
formula

where t denotes the time of hourly MSLP observations (t = 1, 2, . . . , n for the instantaneous hourly MSLP values Pt observed at a site for n consecutive hours). In other words, a cyclone deepening event is defined as an occurrence of a negative MSLP change when at least one of the two relevant instantaneous MSLP values (3 h apart) is below 1000 hPa or, equivalently, an occurrence of δt (note that by definition δt does not exist when both Pt−1 and Pt+2 are above 1000 hPa, so it does not occur at all times t = 1, 2, . . . , n. Also, note that δt are positive values because of the negative sign on the right-hand side of the equation above).

Cyclone deepening rate itself is an index of cyclone activity and has been explored in previous studies (e.g., Gulev et al. 2001; Gyakum et al. 1989; Roebber 1989). Thus, time series of seasonal means, maxima, and variances of cyclone deepening rates were derived for further analyses in addition to the time series of seasonal occurrence counts/frequencies of cyclone deepening events. Note that, due to the use of the 1000-hPa threshold value above, any nonclimatic mean shift in the hourly MSLP data series could still induce an abrupt change in the mean of the time series of frequency/counts of cyclone deepening events and hence affect the estimate of trends in these time series, although it has little effect on the trends estimated using time series of cyclone deepening rates. Therefore, the time series of frequency/counts of cyclone deepening events are subjected to a homogenization procedure before being used for trend analysis (see section 3 below).

The analyses were carried out for each of the four seasons separately, with January–March (JFM), April–June (AMJ), July–September (JAS), and October–December (OND) being defined as winter, spring, summer, and autumn, respectively (such a definition allows us to use all years of data available, whereas defining DJF as winter would result in one less season of data to use and, hence, was not chosen here).

For each of the four seasons, the 90th percentile of cyclone deepening rates δt, denoted as δ90, was also derived. An occurrence of δt > δ90 is referred to as an occurrence of the 90th percentile extreme cyclone deepening events. Time series of seasonal occurrence counts/frequencies of the extreme cyclone deepening events were then derived and subjected to homogenization and trend analysis.

3. Regression analyses and data homogenization

For time series of conventional data such as seasonal means, maxima, and variances of cyclone deepening rates, the linear trend estimate in this study is based on the least squares fit of the following conventional linear regression model:

 
formula

where ɛt denotes a zero mean white noise process. A Student's t test was carried out to check whether or not the slope (trend) β is statistically different from zero [i.e., to assess the statistical significance of the estimated trend (von Storch and Zwiers 1999].

However, assessment of the trend in occurrence frequency of cyclone activity is also the subject of the present study. Since occurrence counts of any type of event over a period (say st in a period of mt observations) have binomial distributions, it is necessary to transform the count (or equivalently frequency) data to map nonnegative counts st ∈ {0, 1, . . . , mt}, or the unit interval of frequency πt = st/mt ∈ [0, 1], onto the whole real line (−∞, +∞) so that conventional regression models can be used to estimate trends in the count or frequency time series.

Mainly for computational convenience, in this study we have chosen the empirical logistic transformation (Hanesiak and Wang 2005; McCullagh and Nelder 1989):

 
formula

Here the expected value of η(t), denoted as 〈η〉, is an approximately unbiased estimate of the log odds λt = log[πt/(1 − πt)] (McCullagh and Nelder 1989). Thus, ηt are called empirical log-odds (also referred to as log-odds hereafter). As clearly shown in Hanesiak and Wang (2005), this transformation does not change the sign or direction of trend, but the linear trend in the time series of log-odds corresponds to an exponential trend in the frequency/count time series; the latter is nearly linear over the observation period of around 50 years. Note that the use of log-odds (or frequency) also greatly diminishes the effect of missing observations (if any), while trend analysis on counts of event occurrence implicitly takes missing observations as no occurrence of event and hence could bias the estimate of trend.

Since mean shifts could exist in the time series of counts (and hence log-odds) due to possible nonclimatic abrupt changes in the mean of hourly MSLP data series (see section 2 above), it is necessary to diminish the effect of any nonclimatic changes on the trend estimate for time series of log-odds (to ensure time series homogeneity) so that a realistic estimate of climatic trend can be obtained. In this study, the following two-phase regression model (Wang 2003)

 
formula

was used to detect a possible mean shift at time tc in time series ηt for the period from N1 to N2 (1 ≤ N1 < N2N). The position of mean shift, tc, and its statistical significance were determined by comparing the sum of squared errors (SSE) of model (4) with that of the null model (3) above. The comparison was done for each and every trial value of tc ∈ {N1 + Nmin, N1 + Nmin + 1, . . . , N2Nmin}, where Nmin is a selected minimum length of segment. The time tc that is associated with the maximum, and statistically significant, reduction in the SSE of model (4) (among all trial values of tc) is chosen as a possible point of the mean shift. The above detection procedure goes on until all possible points of the mean shift are identified or the time series segments are too short to be divided further. Then, metadata (station history and inspection reports) were used to check the veracity of the mean shifts detected statistically. Readers are referred to Wang and Feng (2004), Hanesiak and Wang (2005), and Wang (2003) for more details of the data homogenization technique/procedure and the relevant statistical test.

Once all artificial mean shifts in a time series are detected, we fit the time series with the two-phase regression model (4) or, more precisely, the k-phase regression model (where k is a positive integer, and k > 2 if there are more than one mean shift detected) to estimate the intercept value of each segment, the trend (i.e., slope parameter β) of the time series, and its statistical significance. It can be easily proven that the trend estimate obtained this way is not affected by the artificial mean shifts (see Hanesiak and Wang 2005 for more details).

As described in section 2, an occurrence of δt > δ90 is referred to as an occurrence of the 90th percentile extreme cyclone deepening event. In this study, the above trend analysis was also carried out for the 90th percentile extreme cyclone deepening events. Any trend difference between “all cyclone deepening events” and the 90th percentile extreme cyclone deepening events indicates a change in the probability distribution of cyclone activity.

In addition, we calculated seasonal means and seasonal variances of cyclone deepening rates and estimated linear trends in the time series of seasonal means and variances, separately. Also, we derived 10-yr moving mean time series for the seasonal mean (or variances) time series and then estimated linear trends in these “low pass” filtered time series. As would be expected, while both types of time series share the same sign of trends, trends are more significant in the low-pass filtered time series than in the unfiltered time series (this is because the filtered time series is much less noisy than its unfiltered counterpart). The trends estimated from the low-pass filtered time series, which provide some preliminary knowledge about trends in cyclone activity, are discussed in section 4b.

To show the relationship between the major circulation regimes and cyclone activity in Canada, seasonal mean NAO, PDO, and ENSO indices were used as predictors in a multiple regression analysis to “predict” the seasonal log-odds (or frequencies equivalently) of occurrence of cyclone deepening events at each station. In this study, the seasonal mean NAO, PDO, and ENSO indices were derived from the corresponding monthly indices. The monthly NAO index (available online at http://www.cpc.ncep.noaa.gov/products/precip/CWlink/pna/nao_index.html) was constructed by projecting the monthly mean 500-mb geopotential height anomalies over the Northern Hemisphere onto the loading pattern of the NAO, which was defined as the first leading mode of the rotated empirical orthogonal function (REOF) analysis of monthly mean 500-mb height during the 1950–2000 period. The monthly PDO index (available online at http://jisao.washington.edu/pdo/PDO.latest) was derived as the leading PC of monthly SST anomalies in the North Pacific Ocean, poleward of 20°N (the monthly mean global average SST anomalies were removed to separate this pattern of variability from any “global warming” signal that may be present in the data; see also Zhang et al. 1997). The monthly ENSO index was defined as the average SST anomalies over 6°N–6°S, 180°–90°W (also called the cold tongue SST anomaly index; available online at http://jisao.washington.edu/data/cti/; the anomalies are with respect to the mean period over 1950–79).

Multiple regression analysis was chosen because it conveniently enables us to account for the possible dependence among the three predictor indices and to evaluate the relative importance of each predictor. Here the statistical significance of each regression parameter (which represents the relationship between each predictor and the predictand), and the order of relative importance of each predictor, were determined by performing likelihood ratio tests in which the sum of squared errors of the full model (with three predictors) was compared with the sum of squared errors of the two-predictor model that excludes the predictor being tested (Johnson and Wichern 1982). There will not be significant improvement in the goodness of fit of the regression model when redundant information is brought into the model. For example, if the predictor being tested is so well correlated with one or both of the other two predictors that it becomes redundant, there will not be significant improvement in the fit of the full model over the two-predictor model, and the predictor will be identified to have no significant relationship to the predictand. In other words, the importance of each predictor is assessed in the presence of the other two predictors. This process is somewhat like the stepwise regression approach.

To remove the possible effects of linear trends on the estimates of the regression relationships, least squares estimates of trends were removed from both the predictand and the predictors time series prior to the regression analysis.

For several selected regions, regression of the areal mean seasonal log-odds of cyclone deepening events on the NAO (or PDO or ENSO) index (without detrending) was also carried out to estimate the proportion of interannual cyclone activity variance that can be explained by the circulation index in question. The results are discussed in section 5 below.

4. Observed trends in cyclone activity

a. Climatology of cyclone activity

Figure 4 shows the patterns of long-term (all years) mean cyclone deepening rates (δt) in the four seasons of year. Clearly, cyclone activity in Canada is strongest along the east coast (south of Labrador, as shown in Fig. 3) and weakest in the high Arctic, while it is moderate over the southern Prairies (Alberta and Saskatchewan) and the west coast. It is also very strong in the Great Lakes area in winter (Fig. 4a). Among the four seasons of year, cyclone activity is much stronger in the cold seasons (winter and autumn) than in the warm seasons (spring and summer). The long-term mean cyclone deepening rates exceed 2.5 hPa (3 h)−1 everywhere over the east coast in the cold seasons (Figs. 4a,d).

Fig. 4.

Long-term means of cyclone deepening rate (δt) in each season. The unit is hPa (3 h)−1.

Fig. 4.

Long-term means of cyclone deepening rate (δt) in each season. The unit is hPa (3 h)−1.

The climatological pattern of winter cyclone activity shown in Fig. 4a is in good qualitative agreement with the pattern of cyclone-tracking density (over Canada) diagnosed mainly from the MSLP field of the ECMWF reanalysis (see Fig. 5b in Hoskins and Hodges 2002). This pattern is also qualitatively consistent with the pattern of climatological winter cyclone frequency in Canada as derived from the NNR MSLP data for 1958–99 (see Fig. 1a in Gulev et al. 2001). They all show a high center of cyclone activity over the Canadian east coast to the Great Lakes area, a secondary high center over the west coast and southern Prairies, and a low activity center over the Canadian Arctic. Such consistency suggests that the cyclone deepening rate used in this study is another proper index to measure cyclone activity and that the climatology of cyclone activity in Canada derived from reanalysis data is most likely realistic.

b. Changes in cyclone deepening rates

Figure 5 shows the linear trends estimated for the 10-yr moving means of cyclone deepening rates in each of the four seasons. In winter, the mean deepening rate has experienced significant decreases in southern Canada (the Great Lakes and the region from the central-southern Prairies to British Columbia), but increases in the zone around 60°N and the northeast coast (Fig. 5a). In spring, the trend pattern is characterized by significant increases in central-northeastern Canada (including the central Prairies) with decreases in the other regions (Fig. 5b). The spring and winter trends are similar in southern Canada, but they are of the opposite sign on the east coast, the central Prairies, and the northwest (Figs. 5a,b). Also, the decrease in the Great Lakes–St. Lawrence valley is weaker in spring than in winter. In summer, the deepening rate has an upward trend in northwestern Canada and along the St. Lawrence valley (southern Quebec; see Fig. 3) with a downward trend in most of the other areas (Fig. 5c). In autumn, the pattern is characterized by significant positive trends on the southwest coast and over the Great Lakes area with significant negative trends in the north and the Prairies (Fig. 5d).

Fig. 5.

Trends in the 10-yr moving means of seasonal mean cyclone deepening rate (seasonal mean δt). The large, medium, and small dots indicate trends of p ≥ 0.95, p = 0.80–0.95, and p < 0.80 [where (1 − p) is the statistical significance]. Orange dots superimposed by a plus sign indicate upward trends in the relevant time series of 10-yr moving means, and blue dots indicate downward trends.

Fig. 5.

Trends in the 10-yr moving means of seasonal mean cyclone deepening rate (seasonal mean δt). The large, medium, and small dots indicate trends of p ≥ 0.95, p = 0.80–0.95, and p < 0.80 [where (1 − p) is the statistical significance]. Orange dots superimposed by a plus sign indicate upward trends in the relevant time series of 10-yr moving means, and blue dots indicate downward trends.

The most striking difference among the four seasons is seen in the Great Lakes area and southwest coast between winter and autumn. In these areas, significant increases were identified in autumn with significant decreases in winter (Figs. 5a,d).

Comparison of our results (cf. Figs. 5 and 6) also reveals that trends in the seasonal variances of cyclone deepening rates are similar to those in the corresponding seasonal means, which indicates that increased means were often accompanied by increased variability. In other words, the observed changes in cyclone deepening rate are not just a simple shift of the location but also a change of the shape of the probability distribution. Increased variability indicates a widened probability distribution and hence increased probability for the occurrence of extreme values, and vice versa.

Fig. 6.

As in Fig. 5 but for trends in the 10-yr moving means of seasonal variances of cyclone deepening rates.

Fig. 6.

As in Fig. 5 but for trends in the 10-yr moving means of seasonal variances of cyclone deepening rates.

A notable difference between changes in the mean and those in the variance is seen in southern Saskatchewan in spring and in the west coast in autumn. Decreased seasonal means were accompanied by increased seasonal variances in southern Saskatchewan in spring (Figs. 5b and 6b), while increased seasonal means were accompanied by decreased seasonal variances in the west coast in autumn (Figs. 5d and 6d).

c. Changes in the frequency, duration, and intensity of cyclone activity

The results of trend analysis on the occurrence frequency of cyclone deepening events (see section 3) are shown in Fig. 7. Among the four seasons of the year, the most significant changes are seen in winter. In this season, cyclone deepening events have become significantly more frequent in northwest-central Canada, but less frequent in the south, especially in the east and southwest coasts (Fig. 7a). This trend pattern is in good qualitative agreement with the pattern of change in the frequency of cyclone activity in Canada as diagnosed from the NNR and ERA-40 MSLP data (cf. Fig. 1; see also Gulev et al. 2001), especially in southern Canada. Time series of the areal mean log-odds of winter cyclone deepening events observed in the 60°–70°N zone (shown in Fig. 2c) and in the east coast area (south of 52°N and east of 69°W; not shown) show statistically significant trends. In the 60°–70°N zone, the time series (Fig. 2c) is consistent and well correlated with the time series of log-odds of winter cyclone occurrence as derived from the ERA-40 and NNR cyclone database of Wang et al. (2006); its correlation with the ERA-40 and NNR time series (cf. Fig. 2a) is 0.283 and 0.258, respectively, which are of at least 5% significance. The consistency between the results obtained from analyzing the in situ and reanalysis data (with different methods) indicates that the above-described changes are most likely real and that the way the 3-hourly MSLP changes are used in this study to measure cyclone activity is appropriate.

Fig. 7.

Trends in the seasonal occurrence frequencies of cyclone deepening events. Orange dots superimposed by a plus sign indicate upward trends (increased frequency), and blue dots indicate downward trends (decreased frequency; see section 3a for details). The large, medium, and small dots indicate trends of p ≥ 0.95, p = 0.80–0.95, and p < 0.80.

Fig. 7.

Trends in the seasonal occurrence frequencies of cyclone deepening events. Orange dots superimposed by a plus sign indicate upward trends (increased frequency), and blue dots indicate downward trends (decreased frequency; see section 3a for details). The large, medium, and small dots indicate trends of p ≥ 0.95, p = 0.80–0.95, and p < 0.80.

In spring, changes are not statistically significant for most areas except that the St. Lawrence valley–Labrador region seems to have experienced less frequent cyclone deepening events (Fig. 7b). In summer, the trend pattern is characterized by significant increases in the number of cyclone deepening events on the east coast with decreases of marginal significance in the Great Lakes area and west coast (Fig. 7c). In autumn, cyclone deepening events are less frequent in the Great Lakes area and central-west Canada (Fig. 7d).

In terms of trends in the frequency of cyclone deepening events, the most notable difference among the four seasons of year is seen for the east coast with significant decreases in winter but increases in summer (Figs. 7a,d). The significant increase in northwest-central Canada (Fig. 7a) is also peculiar to winter.

It would be also interesting to know whether or not trends in the frequency of extreme deepening events are different from those of the “overall cyclone deepening events” (i.e., not separating extreme events from nonextreme events). Therefore, similar trend analyses were also carried out for the seasonal log-odds of the 90th percentile extreme deepening events (i.e., deepening rates exceeding the relevant 90th percentiles; see section 3a for details). The results are shown in Fig. 8. Comparison of these trend patterns with those shown in Fig. 7 does reveal some differences in all four seasons, especially in the areas of decreases. More specifically, the winter decrease in the Great Lakes area and southern Prairies is statistically significant for the 90th percentile extreme deepening events but insignificant for the “overall deepening events,” while the increase in northwest-central Canada is weaker for the 90th percentile extreme deepening events (Figs. 7a and 8a). The autumn 90th percentile extreme deepening events in the region from the Prairies to the northwest were identified to have decreased significantly (Fig. 7d), with a higher level of significance than that shown in Fig. 7d. The 90th percentile extreme deepening events seem to have become more frequent in the northwest in summer (Fig. 8c) and the Great Lakes area in autumn (Fig. 8d), while the corresponding overall deepening events seem to have decreased (Figs. 7c,d).

Fig. 8.

As in Fig. 7 but for trends in the seasonal occurrence count/frequency of the 90th extreme percentile cyclone deepening events.

Fig. 8.

As in Fig. 7 but for trends in the seasonal occurrence count/frequency of the 90th extreme percentile cyclone deepening events.

To check whether or not trends in the occurrence frequency of cyclone deepening events really reflect trends in the occurrence frequency of cyclones, seasonal counts of cyclone spells were calculated and subjected to a linear trend analysis (a cyclone spell was defined as a spell that lasts at least 3 h in which all hourly MSLP are below 1000 hPa and the time of the lowest MSLP was assigned as the time of the cyclone spell). The results (not shown) are very similar to those shown in Fig. 7, which corroborates that trends in the frequency of cyclone deepening events do reflect trends in the frequency of cyclone activity. Note that trends estimated from the counts of cyclone spells are not as reliable as those estimated from the frequencies (or log-odds) of cyclone deepening events, mainly because the former could be affected by any temporal change in the number of missing observations (e.g., cyclone spell counts would decrease when there are fewer observations taken even if there is no climatic change in the occurrence frequency of cyclone).

Furthermore, seasonal means of cyclone spell lengths, and seasonal lowest instantaneous MSLP values, were also derived and subjected to a linear trend analysis. The results are shown in Figs. 9 and 10.

Fig. 9.

As in Fig. 7 but for trends in the seasonal mean length of all cyclone spells in the indicated season.

Fig. 9.

As in Fig. 7 but for trends in the seasonal mean length of all cyclone spells in the indicated season.

Fig. 10.

Trends in the seasonal lowest instantaneous MSLP. Orange dots superimposed by a plus sign indicate an increase of cyclone intensity (a decrease in the lowest instantaneous MSLP), and blue dots indicate a decrease of cyclone intensity (an increase of the lowest instantaneous MSLP). The large, medium, and small dots indicate trends of p ≥ 0.95, p = 0.80–0.95, and p < 0.80.

Fig. 10.

Trends in the seasonal lowest instantaneous MSLP. Orange dots superimposed by a plus sign indicate an increase of cyclone intensity (a decrease in the lowest instantaneous MSLP), and blue dots indicate a decrease of cyclone intensity (an increase of the lowest instantaneous MSLP). The large, medium, and small dots indicate trends of p ≥ 0.95, p = 0.80–0.95, and p < 0.80.

Comparison of Figs. 9 and 7 (or the trend patterns of the frequency of cyclone spells, not shown) reveals that the trend patterns of the cyclone spell length (or duration of cyclone activity) have substantial similarity to those of the frequency of cyclone deepening events, especially in winter. The mean duration of cyclone activity has become significantly longer in the high Arctic in all seasons (Fig. 9). In winter it also has become significantly longer in the central-northern Prairies but shorter in most areas on the east and west coasts (see Fig. 9a). Significant shortenings of the mean cyclone spell length are also seen in the Great Lakes area in spring and autumn (Figs. 9b,d).

As shown in Fig. 10, cyclone activity has become stronger (with decreased lowest instantaneous MSLP) over the high Arctic but weaker (with increased lowest instantaneous MSLP) in southern Canada in all seasons. Cyclone activity has also become stronger in the northwest in winter (Fig. 10a) and in the west and northeast coasts in autumn (Fig. 10d).

Note that trends estimated from the lowest MSLP, or from the frequency or duration of cyclone activity, could be subject to the effect of any MSLP data inhomogeneity, especially in southern Canada (Slonosky and Graham 2005). However, this concern is greatly diminished by the consistency of the results with those obtained from analyzing the homogenized data (shown in Fig. 7) and from analyzing the global ERA-40 or NNR 6-hourly MSLP fields (Wang et al. 2006, 2004); that is, the winter storm track in Canada has shifted northward with more frequent cyclone activity in the lower Canadian Arctic but less frequent activity in the south. Also, our results here are consistent with the reported increases in cold season cyclone activity in the Canadian Arctic (Serreze et al. 1997; Paciorek et al. 2002).

5. Major circulation regimes and cyclone activity in Canada

For each season, the multiple regression analysis described in section 3 was carried out with the predictors (the NAO, PDO, and ENSO indices) leading the predictand for 0, 1, 2, and 3 seasons (i.e., lag = 0, 1, 2, 3; cf. Table 1) separately. The total number of stations at which each predictor was found to be the most important predictor (among the three) is given in Table 1 for each season and each lead time. Regardless of the statistical significance level, the simultaneous NAO is the most important predictor at about 52%–71% of the 83 stations in the cold seasons, while the simultaneous PDO in winter and the simultaneous ENSO in autumn are the least important predictors in the sense that they turned out to be the most important predictor at the fewest stations (Table 1a). In this regard, the autumn ENSO index is the most important predictor for winter cyclone activity in Canada, and the summer NAO index for autumn cyclone activity (see the lag = 1 columns in Table 1a). When the predictors lead the predictand for two–three seasons, the PDO is the most important predictor at about one-third of the 83 stations, and the NAO (the ENSO) more (less) than one-third (Table 1a). In winter the number of stations at which the NAO is identified as the most important predictor is the largest when lag = 0 and the smallest when the NAO leads for one season. It picks up again when the NAO leads for two–three seasons, which is probably a manifestation of the annual cycle because the autumn circulation regimes and surface climate have substantial similarity to those of winter. Among those relationships of at least 5% significance, the relative importance of the PDO for spring cyclone activity increases with the increase of lead time (up to two seasons, Table 1b). In particular, the autumn PDO index is the most important predictor with at least 5% significance for spring cyclone activity at 21 of the 83 stations (or 25% of the stations, Table 1b). Further regression analysis (see below) reveals that such a significant relationship is mainly observed in the southwest (cf. the lag = 2 columns in Table 2). This is not surprising because the PDO is a low-frequency (decadal scale) oscillation characterized by an anomalously strong or weak Aleutian low. The extratropical storm track over the North Pacific–America is closely associated with the strength of the Aleutian low, and so is the cyclone activity in Canada, especially southwestern Canada.

Table 1.

The total number of stations where the NAO, PDO, or ENSO was found to be the most important predictor: (a) regardless of the significance level or (b) with at least 5% significance (i.e., p ≥ 0.95). The proportion of stations identified to have a significant relationship (also called “rejection rate”) is an indicator of field significance; e.g., a rejection rate of 13.3% (10%) would indicate a field significance at the 5% level for a field of 30 (80) spatial degrees of freedom (Livezey and Chen 1983). The field of the 83 stations is very likely to have as many as 30 spatial degrees of freedom. So, a rejection rate of 13% [i.e., 11 stations in section (b) of this table] or higher can be deemed to be of at least 5% field significance.

The total number of stations where the NAO, PDO, or ENSO was found to be the most important predictor: (a) regardless of the significance level or (b) with at least 5% significance (i.e., p ≥ 0.95). The proportion of stations identified to have a significant relationship (also called “rejection rate”) is an indicator of field significance; e.g., a rejection rate of 13.3% (10%) would indicate a field significance at the 5% level for a field of 30 (80) spatial degrees of freedom (Livezey and Chen 1983). The field of the 83 stations is very likely to have as many as 30 spatial degrees of freedom. So, a rejection rate of 13% [i.e., 11 stations in section (b) of this table] or higher can be deemed to be of at least 5% field significance.
The total number of stations where the NAO, PDO, or ENSO was found to be the most important predictor: (a) regardless of the significance level or (b) with at least 5% significance (i.e., p ≥ 0.95). The proportion of stations identified to have a significant relationship (also called “rejection rate”) is an indicator of field significance; e.g., a rejection rate of 13.3% (10%) would indicate a field significance at the 5% level for a field of 30 (80) spatial degrees of freedom (Livezey and Chen 1983). The field of the 83 stations is very likely to have as many as 30 spatial degrees of freedom. So, a rejection rate of 13% [i.e., 11 stations in section (b) of this table] or higher can be deemed to be of at least 5% field significance.
Table 2.

The p values of parameters of multiple regression of seasonal log odds of cyclone deepening events in the selected areas on the NAO, PDO, and ENSO indices [predictors 1, 2, and 3, respectively; (1 − p) is the statistical significance]; and the proportions (Pvar) of the total cyclone activity variance that can be explained by the relevant predictor alone. The p values ≥ 0.950 ( ≥ 0.800) are in bold (underlined). A negative sign here indicates that a negative relationship between the predictand and the predictor. The “order” column gives the order of relative importance of the predictors, e.g., “312” means that predictor 3 (ENSO) is the most important predictor, and predictor 2 (PDO), the least important predictor.

The p values of parameters of multiple regression of seasonal log odds of cyclone deepening events in the selected areas on the NAO, PDO, and ENSO indices [predictors 1, 2, and 3, respectively; (1 − p) is the statistical significance]; and the proportions (Pvar) of the total cyclone activity variance that can be explained by the relevant predictor alone. The p values ≥ 0.950 ( ≥ 0.800) are in bold (underlined). A negative sign here indicates that a negative relationship between the predictand and the predictor. The “order” column gives the order of relative importance of the predictors, e.g., “312” means that predictor 3 (ENSO) is the most important predictor, and predictor 2 (PDO), the least important predictor.
The p values of parameters of multiple regression of seasonal log odds of cyclone deepening events in the selected areas on the NAO, PDO, and ENSO indices [predictors 1, 2, and 3, respectively; (1 − p) is the statistical significance]; and the proportions (Pvar) of the total cyclone activity variance that can be explained by the relevant predictor alone. The p values ≥ 0.950 ( ≥ 0.800) are in bold (underlined). A negative sign here indicates that a negative relationship between the predictand and the predictor. The “order” column gives the order of relative importance of the predictors, e.g., “312” means that predictor 3 (ENSO) is the most important predictor, and predictor 2 (PDO), the least important predictor.

According to the (4 × 4 × 3) patterns of regression relationship (some of them shown in Figs. 11 –13), the following four areas were selected for further regression analysis: the east coast (south of 52°N and east of 69°W), the Great Lakes (south of 52°N and between 69° and 85°W), the southwest (south of 60°N and west of 85°W), and the 60°–70°N zone (see Fig. 3). In each of these areas, the statistical significance of the multiple regression of the areal mean seasonal log-odds of cyclone deepening events on the simultaneous or preceding seasonal mean NAO/PDO/ENSO indices are given in Table 2. The proportions (Pvar) of the interannual cyclone activity variance that can be explained by the NAO (or PDO or ENSO) index alone are also listed in Table 2.

Fig. 11.

Multiple regression relationships between the seasonal log-odds of occurrence of cyclone deepening events and the simultaneous seasonal mean NAO indices. Orange dots superimposed by a plus sign indicate positive relationships, and blue dots indicate negative relationships. The large, medium, and small dots indicate relationships of p ≥ 0.95, p = 0.80–0.95, and p < 0.80.

Fig. 11.

Multiple regression relationships between the seasonal log-odds of occurrence of cyclone deepening events and the simultaneous seasonal mean NAO indices. Orange dots superimposed by a plus sign indicate positive relationships, and blue dots indicate negative relationships. The large, medium, and small dots indicate relationships of p ≥ 0.95, p = 0.80–0.95, and p < 0.80.

Fig. 13.

As in Fig. 11 but for the multiple regression relationships between the seasonal log-odds occurrence of cyclone deepening events and the simultaneous seasonal mean ENSO indices.

Fig. 13.

As in Fig. 11 but for the multiple regression relationships between the seasonal log-odds occurrence of cyclone deepening events and the simultaneous seasonal mean ENSO indices.

For the NAO index, as shown in Fig. 11, the patterns of the simultaneous relationships in the four annual seasons have substantial similarity. The most notable seasonal variation is seen in the northwest (from the Yukon southeastward to the northern Prairies) where a significant positive relationship between cyclone activity and the NAO is seen in winter, with an insignificant negative relationship seen in summer. Another notable seasonal variation is seen over the southern Prairies between winter–spring and autumn. In this area, a strong positive NAO is associated with more frequent cyclone activity in winter–spring but less frequent in autumn.

The most significant simultaneous relationships between the NAO index and cyclone activity in Canada are seen in winter and the weakest in summer (Fig. 11 and Table 2). In winter the simultaneous relationship is statistically significant almost everywhere across Canada (Fig. 11a). Strong positive relationships are seen in northwest-central Canada (from the Prairies and northern British Columbia to the Canadian Arctic) with strong negative relationships in the east coast–Great Lakes area (Fig. 11a). As can be seen in Table 2, the areal mean log-odds of winter cyclone deepening events in the 60°–70°N zone and in the east coast area are highly significantly related to the simultaneous seasonal mean NAO index [with p > 0.99, where (1 − p) is the significance]. The NAO index explains 31.1% and 40.8% of the total variance of winter cyclone activity in the 60°–70°N zone and the east coast area, respectively. The areal mean log-odds of winter cyclone deepening events in the Great Lakes area and the southwest also have a marginally significant (0.800 ≤ p < 0.95) relationship with the simultaneous NAO index (Table 2). These indicate that a strong positive NAO is associated with more frequent winter cyclone activity in the western and central Canada but less frequent winter cyclone activity in the east coast–Great Lakes area.

In addition to the above simultaneous relationships, winter cyclone activity in the 60°–70°N zone was also found to have a significant positive relationship with the preceding summer NAO index (i.e., lag = 2; Pvar = 10.9%); in the southwest, it has a marginally significant negative relationship with the preceding spring NAO index (i.e., lag = 3; Pvar = 3.7%; cf. Table 2).

In spring the simultaneous relationship is much weaker in the northwest and the east coast area; and it gets weaker almost everywhere across the country in summer (Figs. 11b and 11c). There is no statistically significant simultaneous relationship between the NAO and summer cyclone activity in the four selected areas (Table 2). However, spring cyclone activity in the 60°–70°N zone and the southwest has a negative relationship with the preceding winter NAO index (i.e., lag = 1), which accounts for 13.8% and 6.7% of the total spring cyclone activity variance, respectively. In the east coast area, spring cyclone activity is also negatively related to the preceding autumn and summer NAO indices (i.e., lag = 2 and 3), which explains 14.7% and 6.5%, respectively, of the total spring cyclone activity variance (Table 2). With a marginal significance, spring cyclone activity in the Great Lakes area is negatively correlated with both the simultaneous and the preceding autumn NAO index (Pvar = 6.5% and 5.8%, respectively), while summer cyclone activity in this area is positively related to the preceding spring NAO index (i.e., lag = 1; Pvar = 5.4%). There is also a marginally significant negative relationship between summer cyclone activity in the southwest and the preceding autumn NAO index (i.e., lag = 1; Pvar = 3.7%).

In autumn cyclone activity in the east coast and the Great Lakes area was found to have a significant negative relationship with the simultaneous seasonal mean NAO index, which accounts for 44.4% and 17.4% of the total autumn cyclone activity variance, respectively (cf. Fig. 11d and Table 2). However, there is no statistically significant relationship between the NAO and autumn cyclone activity in the east coast–Great Lakes area when the NAO leads for one–three seasons (Table 2). Autumn cyclone activity in the southwest has a significant (p = 0.964) negative relationship with the preceding spring NAO index (i.e., lag = 2; p = 0.964, and Pvar = 8.9%); it also seems to be positively related to the preceding summer NAO index (p = 0.887; Pvar = 5.9%). In other words, a strong negative NAO in spring or positive NAO in summer would be followed by more frequent cyclone activity in the southwest in the following autumn. The relationship between the NAO and autumn cyclone activity in the 60°–70°N zone is weak in general; it is only marginally significant when the NAO index leads for one season (p = 0.857 and Pvar = 4.0%; cf. Table 2).

As shown in Fig. 12, the relationship between the frequencies (or log-odds) of cyclone deepening events and the simultaneous PDO indices is relatively weaker in general. Nevertheless, it does show some statistical significance: Strong negative relationships are seen in the region from southern Manitoba–Saskatchewan to northern British Columbia in autumn with positive relationships in the east in spring and summer (Figs. 12c,d). Generally speaking, the lagged relationships between the PDO and cyclone activity in Canada are more significant than the simultaneous ones. In terms of areal mean, the strongest relationships are seen in the southwest, the east coast, and the Great Lakes areas. In spring, cyclone activity in the southwest is positively related to the PDO index two and three seasons earlier (i.e., lag = 2 and 3; Pvar = 14.6% and 5.6%, respectively); while in autumn it is negatively correlated with the simultaneous PDO index (p = 0.974; Pvar = 10.2%). Summer cyclone activity in the southwest also has a positive relationship with the preceding winter PDO index (lag = 2; Pvar = 2.0%). Thus, in the southwest more frequent spring/summer cyclone activity can be expected when a strong positive PDO was observed two seasons earlier. In the east coast area, in both spring and summer, cyclone activity is positively correlated with the PDO index one season earlier (i.e., lag = 1; Pvar = 7.2% for spring and 11.4% for summer). In other words, a strong positive PDO in winter or spring would be followed by more frequent cyclone activity on the east coast in the next season. In the Great Lakes area a strong positive PDO in winter could also be followed by more frequent cyclone activity in the following spring and less frequent cyclone activity in the following autumn (Pvar = 1.2% and 2.8%, respectively). Besides, summer cyclone activity in the east coast area seems to be positively related with the simultaneous PDO index (p = 0.902; Pvar = 8.8%). Autumn and winter cyclone activity in the 60°–70°N zone also seems to have marginally significant relationships to the PDO index three seasons earlier (Table 2).

Fig. 12.

As in Fig. 11 but for the multiple regression relationships between the seasonal log-odds occurrence of cyclone deepening events and the simultaneous seasonal mean PDO indices.

Fig. 12.

As in Fig. 11 but for the multiple regression relationships between the seasonal log-odds occurrence of cyclone deepening events and the simultaneous seasonal mean PDO indices.

As shown in Fig. 13, the relationship between the frequencies of cyclone deepening events and the simultaneous ENSO indices is even weaker (weakest among the three predictors). The relationship is statistically significant only in southwestern British Columbia (the southwest coast) and Ontario–Quebec in winter and spring (Figs. 13a,b). The most significant relationship between the ENSO index and cyclone activity in Canada is seen in the southwest with the ENSO index leading for two seasons; however, the proportion of variance explained is very small (Pvar = 0.3%; cf. Table 2). In terms of areal mean, summer cyclone activity in the east coast seems to be positively related to the ENSO index two and three seasons earlier (Pvar = 5.6% and 10.0%, respectively); while autumn cyclone activity in the 60°–70°N zone seems to have a positive relationship with the simultaneous ENSO index (Pvar = 4.4%). There are other relationships of marginal significance, but accounting for small proportions of the total variance. For example, spring cyclone activity in the east coast–Great Lakes area seems to be negatively correlated with the simultaneous and preceding winter ENSO index, which accounts for less than 3% of the total variance (cf. Table 2).

The relationships between cyclone activity in Canada and the three major circulation regimes are qualitatively consistent with the related reports in previous studies (e.g., Bonsal et al. 2001; Shabbar and Barnston 1996; Shabbar et al. 1997).

6. Summary and discussions

Aiming to assess the climate and trends of cyclone activity in Canada, and to check the veracity of changes in cyclone activity derived from global reanalysis data, we have carried out logistic regression and linear regression analyses to estimate trends in cyclone deepening rates, in the frequency of cyclone activity, and in the seasonal lowest instantaneous MSLP (i.e., seasonal extreme cyclone intensity) after a brief description of the climate of cyclone activity in Canada.

The results show that the most significant changes are seen in winter. In this season, cyclone activity has become significantly more frequent, more durable, and stronger in the lower Canadian Arctic, but less frequent and weaker in the south, especially on the southern east and west coasts; cyclone deepening rates have increased in the zone around 60°N but decreased in the Great Lakes area and southern Prairie–British Columbia region. Extreme winter cyclone activity seems to have experienced a weaker increase in northwest-central Canada but a stronger decline in the Great Lakes area and southern Prairies. These changes were associated with a northward shift of the winter storm track in Canada (Wang et al. 2004) and with warming in the high latitudes. Based on climate projections made with the Canadian coupled climate model, Wang et al. (2004) showed that warmer climates are associated with more strong cyclones in the high Canadian Arctic and Baffin Bay. Using the surface temperature and precipitation data observed in Canada during 1950–98, Zhang et al. (2000) revealed significant increases in surface temperature in northwestern Canada and in precipitation amount in the high Canadian Arctic. All of these support the speculation of Lambert (2004); that is, warmer surface temperatures increase evaporation and thereby increase precipitation and release of latent heat, which contributes to the deepening of cyclones. The consistency of our results with previous results using different data and analysis methods suggests that at least the direction of trend is very likely real and that the cyclone deepening rate used in this study is another proper index to represent cyclone activity.

In addition to the changes in winter, other notable changes include more frequent summer cyclone activity with slower deepening rates on the east coast and less frequent cyclone activity with faster deepening rates in the Great Lakes area in autumn.

The frequency of cyclone activity in Canada was found to be significantly related to the NAO, PDO, and ENSO indices. In particular, a strong positive NAO is always associated simultaneously with more frequent cyclone activity in the high Arctic and less frequent activity on the east coast, regardless of season. The relationships are most significant in winter. Strong positive NAO is also associated with more frequent cyclone activity over the Prairies in winter–spring with less frequent autumn cyclone activity in the Great Lakes area. While strong warm (positive) PDO is simultaneously associated with more frequent cyclone activity on the east coast in summer and the west coast and southeast coast in spring, but less frequent cyclone activity in the Prairies in autumn–winter. The simultaneous relationships between the ENSO and cyclone activity in Canada are weak in general, showing only some marginal significance in the west coast and Quebec–Ontario in winter, in the east coast area in spring, and in the 60°–70°N zone in autumn.

Some significant lagged relationships were also found between the indices of major circulations regimes and cyclone activity in Canada with the indices leading for one–three seasons. In the 60°–70°N zone, in particular, a strong negative winter NAO would be followed by more frequent cyclone activity in the next spring (Pvar = 13.8%), and a strong positive summer NAO, by more frequent cyclone activity in the next winter (Pvar = 10.9%; cf. Table 2). Also, more frequent cyclone activity in the east coast area in spring or in the southwest in autumn can be expected when a strong negative NAO was observed two seasons earlier (Pvar = 14.7% and 8.9%, respectively). As to the PDO, in the east coast area, a strong positive PDO in winter or spring would be followed by more frequent cyclone activity in the next season (Pvar = 7.2% and 11.4%, respectively). While in the southwest, more frequent spring cyclone activity can be expected when a strong positive PDO was observed two seasons earlier (Pvar = 14.6%). Also, more frequent summer cyclone activity in the east coast could be expected when a strong positive ENSO was observed three seasons earlier (Pvar = 10.0%). All these “lagged” relationships might be useful for seasonal forecasting.

Acknowledgments

The authors are grateful to Mr. Amir Shabbar and Dr. Bin Yu for their helpful comments on an earlier version of this manuscript. The anonymous reviewers are also acknowledged for their helpful comments.

REFERENCES

REFERENCES
Blender
,
R.
,
K.
Fraedrich
, and
F.
Lunkeit
,
1997
:
Identification of cyclone-track regimes in the North Atlantic.
Quart. J. Roy. Meteor. Soc
,
123
,
727
741
.
Bonsal
,
B. R.
,
A.
Shabbar
, and
K.
Higuchi
,
2001
:
Impacts of low frequency variability modes on Canadian winter temperature.
Int. J. Climatol
,
21
,
95
108
.
Fyfe
,
J. C.
,
2003
:
Extratropical Southern Hemisphere cyclones: Harbingers of climate change?
J. Climate
,
16
,
2802
2805
.
Graham
,
N. E.
, and
H. F.
Diaz
,
2001
:
Evidence for intensification of North Pacific winter cyclones since 1948.
Bull. Amer. Meteor. Soc
,
82
,
1869
1893
.
Gulev
,
S. K.
,
O.
Zolina
, and
S.
Grigoriev
,
2001
:
Extratropical cyclone variability in the Northern Hemisphere winter from the NCEP/NCAR reanalysis data.
Climate Dyn
,
17
,
795
809
.
Gyakum
,
J. R.
,
J. R.
Anderson
,
R. H.
Grumm
, and
E. L.
Gruner
,
1989
:
North Pacific cold-season surface cyclone activity: 1975–1983.
Mon. Wea. Rev
,
117
,
1141
1145
.
Hanesiak
,
J. M.
, and
X. L.
Wang
,
2005
:
Adverse weather trends in the Canadian Arctic.
J. Climate
,
18
,
3140
3156
.
Higuchi
,
K.
,
C. W.
Yuen
, and
A.
Shabbar
,
2000
:
Ice Storm '98 in Southcentral Canada and Northeastern United States: A climatological perspective.
Theor. Appl. Climatol
,
66
,
61
79
.
Hodges
,
K. I.
,
B. J.
Hoskins
,
J.
Boyle
, and
C.
Thorncroft
,
2003
:
A comparison of recent reanalysis datasets using objective feature tracking: Storm tracks and tropical easterly waves.
Mon. Wea. Rev
,
131
,
2012
2037
.
Hoskins
,
B. J.
, and
K. I.
Hodges
,
2002
:
New perspectives on the Northern Hemisphere winter storm tracks.
J. Atmos. Sci
,
59
,
1041
1061
.
Huang
,
J-P.
,
K.
Higuchi
, and
A.
Shabbar
,
1998
:
The relationship between the North Atlantic Oscillation and El Niño–Southern Oscillation.
Geophys. Res. Lett
,
25
,
2707
2710
.
Hurrell
,
J. W.
,
1995
:
Decadal trends in the North Atlantic Oscillation: Regional temperatures and precipitation.
Science
,
269
,
676
679
.
Johnson
,
R. A.
, and
D. W.
Wichern
,
1982
:
Applied Multivariate Statistical Analysis.
Prentice-Hall, 594 pp
.
Kalnay
,
E.
, and
Coauthors
,
1996
:
The NCEP/NCAR 40-Year Reanalysis Project.
Bull. Amer. Meteor. Soc
,
77
,
437
471
.
Lambert
,
S. J.
,
1996
:
Intense extratropical northern hemisphere winter cyclone events: 1899–1991.
J. Geophys. Res
,
101
,
21319
21325
.
Lambert
,
S. J.
,
2004
:
Changes in winter cyclone frequencies and strengths in transient enhanced greenhouse warming simulations using two coupled climate models.
Atmos.–Ocean
,
42
,
173
181
.
Leckebusch
,
G. C.
, and
U.
Ulbrich
,
2004
:
On the relationship between cyclone and extreme windstorm events over Europe under climate change.
Global Planet. Change
,
44
,
181
193
.
Livezey
,
R. E.
, and
W. Y.
Chen
,
1983
:
Statistical field significance and its determination by Monte Carlo techniques.
Mon. Wea. Rev
,
111
,
46
59
.
McCullagh
,
P.
, and
J. A.
Nelder
,
1989
:
Generalized Linear Models.
2d ed. Chapman & Hall, 511 pp
.
Paciorek
,
C. J.
,
J. S.
Risbey
,
V.
Ventura
, and
R. D.
Rosen
,
2002
:
Multiple indices of Northern Hemisphere cyclone activity, winters 1949–99.
J. Climate
,
15
,
1573
1590
.
Raible
,
C. C.
, and
R.
Blender
,
2004
:
Northern Hemisphere midlatitude cyclone variability in GCM simulations with different ocean representations.
Climate Dyn
,
22
,
239
248
.
Roebber
,
P. J.
,
1989
:
On the statistical analysis of cyclone deepening rates.
Mon. Wea. Rev
,
117
,
2293
2298
.
Rogers
,
J. C.
,
1997
:
North Atlantic storm track variability and its association to the North Atlantic Oscillation and climate variability of northern Europe.
J. Climate
,
10
,
1635
1647
.
Schubert
,
M.
,
J.
Perlwitz
,
R.
Blender
, and
K.
Fraedrich
,
1998
:
North Atlantic cyclones in CO2-induced warm climate simulations: Frequency, intensity, and tracks.
Climate Dyn
,
14
,
827
837
.
Serreze
,
M. C.
,
1995
:
Climatological aspects of cyclone development and decay in the Arctic.
Atmos.–Ocean
,
33
,
1
23
.
Serreze
,
M. C.
,
J. E.
Box
,
R. G.
Barry
, and
J. E.
Walsh
,
1993
:
Characteristics of Arctic synoptic activity, 1952–1989.
Meteor. Atmos. Phys
,
51
,
147
164
.
Serreze
,
M. C.
,
F.
Carse
,
R. G.
Barry
, and
J. C.
Rogers
,
1997
:
Icelandic low cyclone activity: Climatological features, linkages with the NAO, and relationships with recent changes in the Northern Hemisphere circulation.
J. Climate
,
10
,
453
464
.
Shabbar
,
A.
, and
A.
Barnston
,
1996
:
Skill of seasonal climate forecasts in Canada using canonical correlation analysis.
Mon. Wea. Rev
,
124
,
2370
2385
.
Shabbar
,
A.
,
B.
Bonsal
, and
M.
Khandekar
,
1997
:
Canadian precipitation patterns associated with the Southern Oscillation.
J. Climate
,
10
,
3016
3027
.
Sickmoller
,
M.
,
R.
Blender
, and
K.
Fraedrich
,
2000
:
Observed winter cyclone tracks in the Northern Hemisphere in re-analysed ECMWF data.
Quart. J. Roy. Meteor. Soc
,
126
,
591
620
.
Simmonds
,
I.
, and
K.
Keay
,
2000
:
Variability of Southern Hemisphere extratropical cyclone behavior, 1958–97.
J. Climate
,
13
,
550
561
.
Slonosky
,
V. C.
, and
E.
Graham
,
2005
:
Canadian pressure observations and circulation variability: Links to air temperature.
Int. J. Climatol
,
25
.
doi:10.1002/joc.1191
.
Uppala
,
S.
,
2001
:
ECMWF Re-Analysis, 1957–2001. Proc. ECMWF Workshop on Reanalysis, ERA-40 Project Rep. Series, No. 3, Reading, United Kingdom, ECMWF, 1–10
.
von Storch
,
H.
, and
F. W.
Zwiers
,
1999
:
Statistical Analysis in Climate Research.
Cambridge University Press, 484 pp
.
Wang
,
X. L.
,
2003
:
Comments on “Detection of undocumented changepoints: A revision of the two-phase regression model.”.
J. Climate
,
16
,
3383
3385
.
Wang
,
X. L.
, and
Y.
Feng
,
cited
.
2004
:
RHTest user manual. [Available online at http://cccma.seos.uvic.ca/ETCCDMI/RHTest/RHTestUserManual.doc.]
.
Wang
,
X. L.
,
V. R.
Swail
, and
F. W.
Zwiers
,
2004
:
Changes in extra-tropical storm tracks and cyclone activity as derived from two global reanalyses and the Canadian CGCM2 projections of future climate. Preprints, Eighth Int. Workshop on Wave Hindcasting and Forecasting, Oahu, HI, Environment Canada, Paper B1
.
Wang
,
X. L.
,
V. R.
Swail
, and
F. W.
Zwiers
,
2006
:
Climatology and changes of extratropical cyclone activity: Comparison of ERA-40 with NCEP–NCAR reanalysis for 1958–2001.
J. Climate
,
, in press
.
Zhang
,
X.
,
L. A.
Vincent
,
W. D.
Hogg
, and
A.
Niitsoo
,
2000
:
Temperature and precipitation trends in Canada during the 20th century.
Atmos.–Ocean
,
38
,
3
.
395
429
.
Zhang
,
X.
,
J. E.
Walsh
,
J.
Zhang
,
U. S.
Bhatt
, and
M.
Ikeda
,
2004
:
Climatology and interannual variability of Arctic cyclone activity: 1948–2002.
J. Climate
,
17
,
2300
2317
.
Zhang
,
Y.
,
J. M.
Wallace
, and
D. S.
Battisti
,
1997
:
ENSO-like interdecadal variability: 1900–93.
J. Climate
,
10
,
1004
1020
.
Zolina
,
O.
, and
S. K.
Gulev
,
2002
:
Improving the accuracy of mapping cyclone numbers and frequencies.
Mon. Wea. Rev
,
130
,
748
759
.

Footnotes

Corresponding author address: Dr. Xiaolan L. Wang, Climate Research Branch, Meteorological Service of Canada, 4905 Dufferin St., Downsview ON M3H 5T4, Canada. Email: Xiaolan.Wang@ec.gc.ca