Abstract

Seasonal prediction of growing-season start of warm-season crops (GSSWC) is an important task for the agriculture sector to identify risks and opportunities in advance. On the basis of observational daily surface air temperature at 210 stations across Canada, this study found that the GSSWC in most Canadian areas begins during May–June and exhibits significant year-to-year variations that are dominated by two distinct leading empirical orthogonal function modes. The first mode accounts for 20.2% of the total GSSWC variances and features a monosign pattern with the maximum anomalies in central-southern Canada. It indicates that warm-season crops in most Canadian areas usually experience a consistent early or late growing-season start and those in central-southern Canada have the most pronounced interannual variations. The second mode explains 10.8% of the total variances and bears a zonal seesaw pattern in general, accompanied by prominent anomalies covering the west coast of Canada and anomalies with a reverse sign prevailing in central-eastern Canada. Therefore, a strong second-mode year represents an early GSSWC in western Canada and a late GSSWC in the rest of the regions. The predictability sources for the two distinct leading modes show considerable differences. The first mode is closely linked with the North American continental-scale snow cover anomalies and sea surface temperature anomalies (SSTAs) in the North Pacific and Indian Oceans in the prior April. For the second mode, the preceding April snow cover anomalies over western North America and SSTAs in the equatorial-eastern Pacific, North Pacific, and equatorial Indian Oceans provide precursory conditions. These snow cover anomalies and SSTAs sustain from April through May–June, influence the large-scale atmospheric circulation anomalies during the crops’ growing-start season, and contribute to the occurrence of the two leading modes of the GSSWC across Canada. On the basis of these predictors of snow cover anomalies and SSTAs in the prior April, an empirical model is established for predicting the two principal components (PCs) of the GSSWC across Canada. Hindcasting is performed for the 1972–2007 period with a leaving-nine-out cross-validation strategy and shows a statistically significant prediction skill. The correlation coefficient between the observation and the hindcast is 0.54 for PC1 and 0.48 for PC2, both exceeding the 95% confidence level. Because all of these predictors can be readily monitored in real time, this empirical model provides a new prediction tool for agrometeorological events across Canada.

1. Introduction

How to improve seasonal prediction skill of agrometeorological conditions in Canada is becoming an urgent issue and has been receiving fervent research interest during the past decades. Under a global-warming background, climate in Canada is experiencing a dramatic change (e.g., Zhang et al. 2000; Shabbar and Bonsal 2003; Vincent and Mekis 2006). For example, the average increase in annual mean temperatures in southern Canada is 0.9°C since 1895 and winter and spring are warming more than summer and autumn (Vincent and Mekis 2006; Qian et al. 2010). These changes are inevitably modifying Canadian agrometeorological conditions. In light of this, a useful prediction of year-to-year variations of agrometeorological conditions in Canada will not only benefit Canadian agriculture but will also enhance Canadian preparedness and adaptation to global climate change. The amount of research on seasonal prediction of agrometeorological conditions in Canada has unfortunately been relatively small up to now. This motivates us to conduct this work.

Agrometeorological conditions include quite a few aspects, and several indices have been proposed to measure their variations (e.g., Vincent and Mekis 2006; Qian et al. 2010). Among them, growing-season start of warm-season crops (GSSWC) is a principal one. Warm-season crops include bean, corn, pea, soybean, and so on. It was found that warm-season crops in most Canadian areas will not start growing until daily mean surface air temperature (Ts) exceeds 10°C for 10 consecutive days. Therefore, seasonal prediction of the Ts associated with the GSSWC is practically an issue of predicting when a period of 10 consecutive days with mean Ts reaching 10°C emerges in a year.

It has long been recognized that the physical basis of seasonal prediction of climate events lies in coupled mechanisms between atmosphere and low boundary forcing anomalies such as snow cover and sea surface temperature anomalies (SSTAs) (e.g., Charney and Shukla 1981; Shukla 1998) because the atmosphere, on its own, lacks the mechanisms to generate predictable variations beyond two weeks (Lorenz 1963). Previous studies revealed that the interannual variations of seasonal mean Ts (winter in particular) in Canada are greatly influenced by the surrounding ocean. For example, El Niño–Southern Oscillation (ENSO) is a primary predictability source for interannual variations of the winter climate (e.g., Ropelewski and Halpert 1986; Hurrell 1996; Shabbar and Khandekar 1996; Shabbar and Barnston 1996; Wang et al. 2000). The influence of ENSO extends to Canada through atmospheric teleconnections related to tropical diabatic forcing (e.g., Horel and Wallace 1981; Trenberth 1990; Lin and Derome 2004; Lin et al. 2005). Besides ENSO, SSTAs in the Indian Ocean may also contribute to seasonal prediction of winter Ts over North America (e.g., Wu et al. 2009a; Lin and Wu 2011).

Because the ocean accounts for only a portion of Ts variability (Ting et al. 1996; Hurrell 1996), the substantial landmass is a viable candidate for at least amplifying climate anomalies. Snow cover is the most variable land surface condition in both time and space and exerts profound influences on winter Ts variations in the Northern Hemisphere (Robinson et al. 1993; Cohen 1994; Gutzler and Rosen 1992; Wang et al. 2010). For instance, Foster et al. (1983) investigated the relationships between snow cover and temperature over North America and Eurasia. Barnett et al. (1988) discussed the effect of Eurasian snow cover on global climate. Lin and Wu (2011) revealed that the prior autumn snow cover anomalies over the Tibetan Plateau can sustain through the ensuing winter and exert profound influences on winter Ts over Canada.

Although many studies have been conducted on interannual variations of Ts and its seasonal prediction, most of them focused on boreal winter season and few focused on Ts in transitional seasons (late spring–early summer in particular). The latter is directly connected with the GSSWC across Canada. In this study, we attempt to answer the following questions: What are the major features of the GSSWC across Canada? How does the GSSWC link to the large-scale atmospheric circulations and the low boundary forcing (SSTAs and snow cover anomalies)? What are the predictors for the GSSWC if it is predictable, and how do they contribute to seasonal prediction of the GSSWC from the Ts perspective?

The outline of this study is as follows. Section 2 introduces the datasets and method used in this study. Section 3 suggests that the air temperature conditions favorable for the GSSWC in most Canadian areas begin in May–June and are dominated by two distinct leading modes. Section 4 presents the large-scale three-dimensional circulation features associated with the two distinct modes and predictability sources for them. In section 5, an empirical model is established to predict the principal components (PCs) of the two distinct modes based on the prior April snow cover anomalies and SSTAs. Hindcasting is performed for the 1972–2007 period. The last section summarizes major findings and discusses some outstanding issues.

2. Data and method

The main datasets employed in this study include 1) the homogenized Canadian historical daily Ts at 210 relatively evenly distributed stations across Canada (Vincent et al. 2002; see Fig. 1); 2) the 40-yr European Centre for Medium-Range Weather Forecasts Re-Analysis (ERA-40; Uppala et al. 2005) and the ERA-Interim reanalysis data; 3) the Met Office Hadley Centre’s SST datasets gridded at 1.0° × 1.0° resolution (Rayner et al. 2003); 4) the Northern Hemisphere snow cover data gridded at 2.0° × 2.0° resolution, obtained online (http://www.cpc.ncep.noaa.gov/data/snow/).

Fig. 1.

Distribution of 210 Ts gauge stations across Canada.

Fig. 1.

Distribution of 210 Ts gauge stations across Canada.

The daily Ts data at 210 stations have been adjusted to account for inhomogeneities caused by changes in site exposure, location, instrumentation, observer, and observing procedures. The period of our analysis on the GSSWC covers from 1957 through 2007. To get a longer time length covering the period from 1957 through 2008, the ERA-40 and ERA-Interim data are combined together. We use the ERA-40 data for the period 1957–2001 and extend the data from 2002 through 2008 by using ERA-Interim data (Wang et al. 2010; Lin and Wu 2011). To maintain temporal homogeneity, the 2002–08 ERA-Interim data were adjusted by removing the climatological difference between the ERA-40 and ERA-Interim data.

Because this study focuses on seasonal prediction of the GSSWC from the Ts perspective, the GSSWC is defined as the beginning date of 10 consecutive days with their daily mean Ts reaching 10°C. The beginning date is represented by the number of days with reference to 1 January. For example, if the beginning date is 2 January, the GSSWC value will be 2. Missing data in GSSWC are replaced by the climatological value at this station. Stations with more than 10% of missing GSSWC (viz., five observations in this study) are excluded from the analysis. To derive the leading modes, we performed an empirical orthogonal function (EOF) analysis on the GSSWC. The EOF analysis was carried out by constructing an area-weighted covariance matrix.

3. Major features of the GSSWC across Canada

Figure 2 presents the climatological GSSWC across Canada. A prominent feature is that the GSSWC value increases with latitude. It indicates that the warm-season crops in high-latitude regions start growing later than those in low-latitude regions. This is consistent with seasonal alternation of the solar radiation. The earliest growing start date of the warm-season crops climatologically begins around 1 May (120 days). The period May–June (MJ) is the essential growing start season for Canadian warm-season crops. The seasonal prediction of the GSSWC in the following section will focus on the MJ period.

Fig. 2.

Climatological GSSWC (gray shading; days). The GSSWC refers to the day numbers with reference to 1 Jan (yearday).

Fig. 2.

Climatological GSSWC (gray shading; days). The GSSWC refers to the day numbers with reference to 1 Jan (yearday).

Figure 3 displays the two leading modes of the GSSWC. The first mode accounts for 20.2% of the total variance (Fig. 3a). According to the rule given by North et al. (1982), the first mode is statistically distinguished from the rest of the eigenvectors in terms of the sampling error bars (not shown). The second mode, which accounts for 10.8% of the total variance (Fig. 3c), is not separable from the rest of the high modes. Nevertheless, the agrometeorological meaning of the first two modes is examined here.

Fig. 3.

(a) Spatial pattern (color shading; days) and (b) the corresponding PC of the first EOF mode of the GSSWC. (c),(d) As in (a) and (b), but for the second mode. The numbers in the parentheses indicate fractional variance of the EOF modes.

Fig. 3.

(a) Spatial pattern (color shading; days) and (b) the corresponding PC of the first EOF mode of the GSSWC. (c),(d) As in (a) and (b), but for the second mode. The numbers in the parentheses indicate fractional variance of the EOF modes.

The first mode basically shows a monosign pattern with maximum loading located in the central-southern Canada, its amplitude decreasing northwestward and northeastward (Fig. 3a). The central-southern Canada has the most significant year-to-year variations. PC1 is primarily dominated by interannual variability (Fig. 3b). In this study, a high (low) PC1 year refers to an early (late) growing-season start year across Canada. An interesting phenomenon is that most of the years before 1975 have a negative PC1; namely, the warm crops in Canada are more likely to have a late growing-season start during the 1957–75 period than they do after 1975. This leads to a decreasing tendency in GSSWC, which is basically consistent with the result from Qian et al. (2010). It is still not clear whether this phenomenon is due to global warming, because after 1980 when the dramatic global warming happens PC1 does not exhibit a significantly increasing tendency in its positive phases as expected.

The prominent feature of the EOF2 mode is a zonal dipole pattern with anomalies of opposite signs in western Canada and the rest of the region except the area east of Hudson Bay (Fig. 3c). The extreme-value centers are located along the west coast of Canada. A high (or low) PC2 year corresponds to a late-west–early-east (early-west–late-east) GSSWC pattern over most Canadian areas. PC2 is also dominated by interannual variability, and its amplitude has increased considerably since 1980, with negative phases in particular (Fig. 3d). It indicates that the early-west–late-east GSSWC patterns are more pronounced than the late-west–early-east GSSWC patterns in the latest 27 years, which is basically consistent with the results in Zhang et al. (2000).

The distinct spatial–temporal structures of the two leading modes imply that they may have different physical origins and predictability sources. In the next section, we will examine the large-scale circulation anomalies accompanied by the two leading modes and their predictability sources.

4. Circulation anomalies and predictability sources

To better understand the linkage between the two distinct modes of the GSSWC and their predictability sources, first we need to examine the simultaneous large-scale circulations associated with these two modes. Figure 4 shows MJ surface circulation anomalies regressed to the two leading PCs along with the climatological values. One prominent feature of the atmospheric circulations near the surface in a high-PC1 MJ is an anomalous Ts warming area prevailing over the North American (NA) continent and centered in central-southern Canada (shadings in Fig. 4b). The Ts pattern resembles well the spatial pattern of the EOF1 mode (Fig. 3a). It indicates an early GSSWC in Canada is often accompanied by a warmer-than-normal MJ, and vice versa. Another prominent feature is one gigantic positive sea level pressure (SLP) anomaly center occupying the entire northeastern Pacific Ocean (contours in Fig. 4b) with significant anticyclonic wind anomalies at 925 hPa (vectors in Fig. 4b). It is located slightly to the north of the climatological Hawaiian high pressure system (Fig. 4a). This pattern reflects a stronger-than-normal and northward-shifted Hawaiian high pressure system. Meanwhile, a negative SLP anomaly center controls the midlatitude western North Atlantic Ocean, which is corresponding to a weaker-than-normal North Atlantic high pressure system. These are favorable for a warmer-than-normal MJ in Canada.

Fig. 4.

For MJ: (a) climatological values in SLP (contours; hPa), Ts (color shading; °C), and 925-hPa winds (vectors; m s−1) and their anomalies regressed to (b) PC1 and (c) PC2.

Fig. 4.

For MJ: (a) climatological values in SLP (contours; hPa), Ts (color shading; °C), and 925-hPa winds (vectors; m s−1) and their anomalies regressed to (b) PC1 and (c) PC2.

For a strong second-mode MJ, a zonal seesaw Ts pattern prevails over the NA continent (shadings in Fig. 4c), with warm Ts anomalies over central-eastern Canada and cold anomalies over western Canada. This pattern also resembles the spatial pattern of the EOF2 mode (Fig. 3c). One large positive SLP anomaly center associated with anticyclonic surface wind anomalies occupies the Aleutian region, which implies a weaker-than-normal Aleutian low pressure system. Northerly surface wind anomalies prevail in the northeastern Pacific that advect cooler and drier air southward from the north, which decreases Ts over the northeastern Pacific–western Canada in a high-PC2 MJ. In a low-PC2 MJ, the situation tends to be opposite.

Figure 5 compares MJ midtroposphere circulation anomalies regressed to the two leading modes along with the climatological values. For a strong first-mode MJ, the NA continent is basically controlled by positive geopotential height H anomalies at 500 hPa centered over central-southern Canada (Fig. 5b), which is primarily above the surface warming center (Fig. 4b). This positive H anomaly extends eastward across the midlatitude North Atlantic, with another center over the eastern Atlantic. The high pressure system in the mid–high troposphere over Canada (Fig. 5b) may lead to clear sky and increased solar radiation and consequently favor a warmer-than-normal MJ in Canada. A salient negative H anomaly center and a positive anomaly center occupy the Aleutian region and the central North Pacific, respectively. Two negative H anomaly centers are located over the west and east coasts of the United States, expanding toward the Pacific and Atlantic, respectively. Because the climatological ridge over the northeastern Pacific and the west coast of the NA continent tilts southeastward from Alaska to the Rocky Mountains (Fig. 5a), the negative H anomalies over the west coast of North America imply an eastward shift of the high ridge toward the central NA continent. This suppresses synoptic eddy activities in Canada. The negative H anomalies over the midwestern Atlantic imply a weakened North Atlantic subtropical high.

Fig. 5.

As in Fig. 4, but for 500-hPa geopotential height H (contours; 0.1 × gpm), and winds (vectors; m s−1).

Fig. 5.

As in Fig. 4, but for 500-hPa geopotential height H (contours; 0.1 × gpm), and winds (vectors; m s−1).

For a strong second-mode MJ, a tremendous positive H anomaly center prevails over the central-eastern NA continent while a negative H anomaly center controls the northwest of Canada and Alaska. An anomalous positive H belt extends from the central North Pacific to the west coast of the United States. This pattern tends to strengthen the synoptic eddy activities over western Canada and weaken those over central-eastern Canada and the United States.

The above circulation anomalies associated with the two distinct modes are likely intimately coupled with the anomalous low boundary conditions such as snow cover and SST; namely, they are potential predictability sources. Figure 6 presents the correlation map between the two PCs and the prior April snow cover over North America. In April of a high-PC1 year, large areas of significantly negative correlations cover most NA continental areas (Fig. 6a). It indicates that a reduced (excessive) NA snow cover in April signifies precursory conditions for an early (late) GSSWC in Canada. Meanwhile, with respect to SST, negative correlations are observed in the North Pacific and positive correlations in the Indian Ocean basin (Fig. 7a), which means that a colder-than-normal North Pacific and a warmer-than-normal Indian Ocean provide preceding signals for an early GSSWC in Canada, and vice versa.

Fig. 6.

Correlation coefficients between North American snow cover in the prior April and GSSWC (a) PC1 and (b) PC2. The interval of contours is 0.1. The shaded regions represent correlation coefficients that exceed the 95% confidence level.

Fig. 6.

Correlation coefficients between North American snow cover in the prior April and GSSWC (a) PC1 and (b) PC2. The interval of contours is 0.1. The shaded regions represent correlation coefficients that exceed the 95% confidence level.

Fig. 7.

Correlation coefficients between SSTAs in the prior April and (a) PC1 and (b) PC2. The interval of contours is 0.1. The shaded regions represent correlation coefficients that exceed the 95% confidence level.

Fig. 7.

Correlation coefficients between SSTAs in the prior April and (a) PC1 and (b) PC2. The interval of contours is 0.1. The shaded regions represent correlation coefficients that exceed the 95% confidence level.

In April of a high-PC2 year, positive correlations with the April snow cover are observed in western Canada, expanding northwestward toward Alaska (Fig. 6b), which means that a high-PC2 mode of GSSWC in Canada is usually preceded by an excessive snow cover in western Canada in April. The anomalously negative SST correlation areas are basically located in the equatorial eastern Pacific, the northeastern Pacific adjacent to the west coast of North America, and the tropical Indian Ocean (Fig. 7b). The signal in the Pacific is reminiscent of a La Niña SSTA. It implies that a colder-than-normal SST in the above-mentioned ocean areas in April are often prior to a high-PC2 mode of GSSWC in Canada.

It is known that the atmosphere responds to an anomalous low boundary forcing within about two weeks, even for a remote response; thus on the seasonal time scale the interaction between the atmosphere and low boundary forcing can be regarded as a simultaneous relationship (e.g., Charney and Shukla 1981; Wu et al. 2009b). Can these prior April low boundary forcing anomalies associated with the two distinct modes persist through the MJ period? Figures 8 and 9 display snow cover anomalies and SSTAs in MJ associated with the two distinct modes. The major feature of the two figures is that both the snow cover and the SST in MJ bear a similar anomaly pattern with their correlation maps in the prior April (Figs. 6 and 7). For a strong first-mode year, reduced snow cover anomalies appear in large areas of the NA continent during MJ (Fig. 8a), whereas colder-than-normal SSTAs emerge in the North Pacific and warmer SSTAs in the Indian Ocean basin (Fig. 9a). For a strong second mode, excessive snow cover anomalies are basically located in western Canada, expanding to Alaska (Fig. 8b), and colder SSTAs are in the equatorial eastern Pacific, northeastern Pacific adjacent to the west coast of North America, and the tropical Indian Ocean (Fig. 9b). These anomalous patterns basically resemble their correlation patterns (Figs. 6 and 7). It manifests that these low boundary forcing anomalies can persist from the prior April through MJ, which makes them predictors for the two distinct modes of GSSWC in Canada.

Fig. 8.

North American MJ snow cover anomalies regressed to (a) PC1 and (b) PC2. The interval of contours is 2%. The shaded regions exceed 95% confidence level.

Fig. 8.

North American MJ snow cover anomalies regressed to (a) PC1 and (b) PC2. The interval of contours is 2%. The shaded regions exceed 95% confidence level.

Fig. 9.

The MJ composite differences in SST between high-PC and low-PC (high minus low) years for (a) PC1 and (b) PC2. A high-PC (low PC) year is measured by 1 standard deviation. The interval of contours is 0.3°C. The shaded regions exceed the 95% confidence level.

Fig. 9.

The MJ composite differences in SST between high-PC and low-PC (high minus low) years for (a) PC1 and (b) PC2. A high-PC (low PC) year is measured by 1 standard deviation. The interval of contours is 0.3°C. The shaded regions exceed the 95% confidence level.

5. Seasonal prediction

To verify how well the above predictors contribute to the seasonal prediction of the GSSWC, an empirical seasonal prediction model is established using a linear-regression method for the period of 1972–2007 [see Eqs. (1) and (2)]:

 
formula
 
formula

where Eqs. (1) and (2) are for PC1 and PC2, respectively. In Eq. (1), x11 denotes the April normalized snow cover averaged over 40°–63°N, 125°–65°W and x12 and x13 refer to April normalized SSTAs averaged in the Indian Ocean (20°S–10°N, 50°–85°E) and the North Pacific (10°–20°N, 180°–140°W plus 37°–49°N, 160°E–150°W), respectively (boxes in Fig. 7a). All of these predictors show an intimate linkage with PC1 (Fig. 10a), with their correlation coefficients being −0.57, 0.34, and −0.48, respectively, all reaching the 95% confidence level. In Eq. (2), x21 denotes the April normalized snow cover averaged over 45°–53°N, 125°–105°W and x22, x23, and x24 refer to April normalized SSTAs averaged in the Indian Ocean (5°S–5°N, 60°–100°E), the North Pacific (45°–60°N, 160°–120°W), and the tropical eastern Pacific (15°S–20°N, 160°–120°W), respectively (boxes in Fig. 7b). All of these predictors show an intimate linkage with PC2 (Fig. 10b), with their correlation coefficients being 0.41, −0.45, −0.61, and −0.49, respectively (beyond the 95% confidence level).

Fig. 10.

Time series of April predictors for (a) PC1 and (b) PC2. The correlation coefficients between the PCs and their predictors are indicated in the parentheses. Note that NP, IO, NA, and TP refer to the North Pacific, the Indian Ocean, North America, and the tropical Pacific, respectively.

Fig. 10.

Time series of April predictors for (a) PC1 and (b) PC2. The correlation coefficients between the PCs and their predictors are indicated in the parentheses. Note that NP, IO, NA, and TP refer to the North Pacific, the Indian Ocean, North America, and the tropical Pacific, respectively.

The cross-validation method is performed to hindcast PC1 and PC2 for the 1972–2007 period (Michaelsen 1987; Wu et al. 2009b). To warrant a robust hindcast, we choose a leaving-nine-out strategy (Blockeel and Struyf 2002). The relevant procedures are as follows: The cross-validation method systematically deletes nine years from the period 1972–2007, derives a forecast model from the remaining years, and tests it on the deleted cases. Note that the choice of “leaving nine out” is not random. Blockeel and Struyf (2002) suggested that randomly choosing 20%–30% of the data to be in a test dataset and the remainder as a training set for performing regression can prevent overfitting or wasting of data. For the two leading PCs, 25% of the whole hindcast period (36 yr) is equal to 9 yr. That is why we choose a leaving-nine-out strategy.

The cross-validated estimates of PCs are shown in Fig. 11. For PC1, the correlation coefficient between the observation (black line in Fig. 11a) and the cross-validated estimates of the empirical scenario (red line in Fig. 11a) reaches 0.54, exceeding the 95% confidence level. For PC2, the correlation coefficient between the observation (black line in Fig. 11b) and the cross-validated estimates of the empirical scenario (red line in Fig. 11b) reaches 0.48, also exceeding the 95% confidence level. Therefore, the empirical method shows a promising hindcast skill. Because all of these predictors can be readily monitored in real time, this empirical model provides a new prediction tool for agroclimatic events in Canada.

Fig. 11.

Comparison of the observed (solid curves) and the hindcast (dot–dashed curves) PCs made by the empirical seasonal prediction model for (a) PC1 and (b) PC2. The correlation coefficients between the observation and the hindcast are indicated in the parentheses.

Fig. 11.

Comparison of the observed (solid curves) and the hindcast (dot–dashed curves) PCs made by the empirical seasonal prediction model for (a) PC1 and (b) PC2. The correlation coefficients between the observation and the hindcast are indicated in the parentheses.

6. Conclusions and discussion

Seasonal prediction of agroclimatic conditions in Canada is of central importance for the Canadian agricultural sector to identify risks and opportunities in advance and has become a focal issue under a global-warming background. This paper focuses on seasonal prediction of the GSSWC from the Ts perspective (Qian et al. 2010). Based on observational daily Ts data at 210 stations across Canada (Vincent et al. 2002), we find that the GSSWC in most Canadian areas climatologically begins in May–June and exhibits significant year-to-year variations that are dominated by two distinct leading modes (North et al. 1982). The first mode accounts for 20.2% of the total GSSWC variances and features a monosign pattern with the maximum anomalies in central-southern Canada. It indicates that warm-season crops in most Canadian areas usually experience a consistent early or late growing-season start while those in central-southern Canada have the most pronounced interannual variations. The second mode explains 10.8% of the total variances and bears a zonal seesaw pattern in general, accompanied by prominent anomalies covering the west coast of Canada and anomalies with reverse sign prevailing in central-eastern Canada. Therefore, a strong second-mode year represents an early GSSWC in western Canada and a late GSSWC in the rest of the region, and vice versa. Seasonal prediction of the two leading modes is essential for seasonal prediction of the GSSWC across Canada.

The predictability sources for the two distinct modes are also examined. The first mode is intimately connected with the North American continental-scale snow cover anomalies and SSTAs in the North Pacific and Indian Oceans in the prior April. For the second mode, the preceding April snow cover anomalies over western North America and SSTAs in the equatorial-eastern Pacific, North Pacific, and equatorial Indian Oceans provide precursory conditions. These low boundary forcing anomalies can persist from April through MJ.

The question arises as to how the April low boundary forcing anomalies affect the ensuing MJ large-scale atmospheric circulations. In April of a high-PC1 year, large areas of reduced NA snow cover in April persist through MJ (Figs. 6a and 8a) and the lowest layers of the atmosphere are warmed because of the low albedo of snow cover (e.g., Foster et al. 1983). Warming at the bottom of the atmospheric column produces convergent flow, weakening high pressure in the region and causing it to expand (contours in Fig. 4b). Meanwhile, the tripole SSTA pattern in the North Pacific (Fig. 9a) may favor a stronger-than-normal and northward-shifted Hawaiian high pressure system and an enhanced Aleutian low pressure system. In a low-PC1 year, the situation is just the opposite. For a high-PC2 (low PC2) year, an excessive (reduced) snow cover in western Canada cooled (warmed) the lowest layers of the atmosphere because of the high (low) albedo of snow cover and induced positive (negative) SLP anomalies over the local region (contours in Fig. 4c). The tremendous H positive (negative) anomaly center with anticyclonic (cyclonic) wind anomalies (Fig. 5c) indicates a Rossby wave response to the La Niña–like (El Niño–like) SSTAs (Fig. 9b) (Hoskins and Karoly 1981; Sardeshmukh and Hoskins 1988). Thus, the snow cover anomalies and SSTAs associated with the two leading modes can interpret well the main features of the corresponding large-scale atmospheric circulations.

On the basis of these predictors of snow cover anomalies and SSTAs in the prior April, we establish an empirical model to predict the PC1 and PC2 of the GSSWC. Hindcasting is performed for the 1972–2007 period with a leaving-nine-out cross-validation strategy and shows a significant prediction skill. The correlation coefficient between the observation and the hindcast is 0.54 for PC1 and 0.48 for PC2, both exceeding the 95% confidence level. In this study we have focused on seasonal prediction of the two leading modes of the GSSWC, which explain about 31% of the seasonal mean variance of the GSSWC variability over Canada. Because the EOF was done over a large area covering the whole country, the percentage of variance explained refers to an average of the whole analysis area. For local regions such as the anomaly centers over southern and western Canada, the variance explained by these two modes would be much larger.

Here, we used the April snow cover and SST to establish this empirical seasonal prediction model for the Canadian GSSWC. By using the latest signals of the predictors, we want to show the best prediction skill of this model. Because the snow cover and SST have a longer memory than 1 month (e.g., Wu et al. 2009b; Lin and Wu 2011), this indicates that the empirical model can use the earlier snow cover and SST signals to do 1-month-lead or 2-month-lead prediction. Because all of these predictors can be readily monitored in real time, this empirical model provides a new prediction tool for agrometeorological events across Canada.

This seasonal prediction model assumes that the two distinct modes are stable on interannual time scales. For predictions on a time scale of a decade or longer, the predictability sources and relevant physical processes are likely to be different from those with the interannual variability. If the two distinct modes change with time—that is, interdecadal changes—then the predictors and the prediction scenario may also change correspondingly. In addition, how can the SSTAs in the Indian Ocean have an impact on circulation anomalies associated with the two distinct modes of the GSSWC and what kind of physical processes are involved? These are still open questions. The hypotheses concerning the origins of the first leading mode call for further numerical and theoretical studies.

Acknowledgments

Zhiwei Wu is supported by the Sustainable Agriculture Environment Systems (SAGES) research initiative of Agriculture and Afri-Food Canada through the Natural Sciences and Engineering Research Council of Canada (NSERC) Fellowship Program.

REFERENCES

REFERENCES
Barnett
,
T. P.
,
L.
Dumenil
,
U.
Schlese
, and
E.
Roeckner
,
1988
:
The effect of Eurasian snow cover on global climate
.
Science
,
239
,
504
507
.
Blockeel
,
H.
, and
J.
Struyf
,
2002
:
Efficient algorithms for decision tree cross-validation
.
J. Mach. Learn. Res.
,
3
,
621
650
.
Charney
,
J. G.
, and
J.
Shukla
,
1981
:
Predictability of monsoons
.
Monsoon Dynamics, J. Lighthill and R. P. Pearce, Eds., Cambridge University Press, 99–109
.
Cohen
,
J.
,
1994
:
Snow cover and climate
.
Weather
,
59
,
150
156
.
Foster
,
J.
,
M.
Owe
, and
A.
Rango
,
1983
:
Snow cover and temperature relationships in North America and Eurasia
.
J. Climate Appl. Meteor.
,
22
,
460
469
.
Gutzler
,
D. S.
, and
R. D.
Rosen
,
1992
:
Interannual variability of wintertime snow cover across the Northern Hemisphere
.
J. Climate
,
5
,
1441
1447
.
Horel
,
J. D.
, and
J. M.
Wallace
,
1981
:
Planetary-scale atmospheric phenomena associated with the Southern Oscillation
.
Mon. Wea. Rev.
,
109
,
813
829
.
Hoskins
,
B. J.
, and
D. J.
Karoly
,
1981
:
The steady linear response of a spherical atmosphere to thermal and orographic forcing
.
J. Atmos. Sci.
,
38
,
1179
1196
.
Hurrell
,
J. W.
,
1996
:
Influence of variations in extratropical wintertime teleconnections on Northern Hemisphere temperature
.
Geophys. Res. Lett.
,
23
,
665
668
.
Lin
,
H.
, and
J.
Derome
,
2004
:
Nonlinearity of extratropical response to tropical forcing
.
J. Climate
,
17
,
2597
2608
.
Lin
,
H.
, and
Z. W.
Wu
,
2011
:
Contribution of the autumn Tibetan Plateau snow cover to seasonal prediction of North American winter temperature
.
J. Climate
,
24
,
2801
2813
.
Lin
,
H.
,
J.
Derome
, and
G.
Brunet
,
2005
:
Tropical Pacific link to the two dominant patterns of atmospheric variability
.
Geophys. Res. Lett.
,
32
,
L03801
,
doi:10.1029/2004GL021495
.
Lorenz
,
E. N.
,
1963
:
Deterministic nonperiodic flow
.
J. Atmos. Sci.
,
20
,
130
141
.
Michaelsen
,
J.
,
1987
:
Cross-validation in statistical climate forecast model
.
J. Climate Appl. Meteor.
,
26
,
1589
1600
.
North
,
G. R.
,
T. L.
Bell
,
R. F.
Cahalan
, and
F. J.
Moeng
,
1982
:
Sampling errors in the estimation of empirical orthogonal functions
.
Mon. Wea. Rev.
,
110
,
699
706
.
Qian
,
B.
,
X.
Zhang
,
K.
Chen
,
Y.
Feng
, and
T.
O’Brien
,
2010
:
Observed long-term trends for agroclimatic conditions in Canada
.
J. Appl. Meteor. Climatol.
,
49
,
604
618
.
Rayner
,
N. A.
,
D. E.
Parker
,
E. B.
Horton
,
C. K.
Folland
,
L. V.
Alexander
,
D. P.
Rowell
,
E. C.
Kent
, and
A.
Kaplan
,
2003
:
Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century
.
J. Geophys. Res.
,
108
,
D14407
,
doi:10.1029/2002JD002670
.
Robinson
,
D. A.
,
K. F.
Dewey
, and
R. R.
Heim
,
1993
:
Global snow cover monitoring: An update
.
Bull. Amer. Meteor. Soc.
,
74
,
1689
1696
.
Ropelewski
,
C. F.
, and
M. S.
Halpert
,
1986
:
North American precipitation and temperature patterns associated with the El Niño/Southern Oscillation (ENSO)
.
Mon. Wea. Rev.
,
114
,
2352
2362
.
Sardeshmukh
,
P. D.
, and
B. J.
Hoskins
,
1988
:
The generation of global rotational flow by steady idealized tropical divergence
.
J. Atmos. Sci.
,
45
,
1228
1251
.
Shabbar
,
A.
, and
A. G.
Barnston
,
1996
:
Skill of seasonal climate forecasts in Canada using canonical correlation analysis
.
Mon. Wea. Rev.
,
124
,
2370
2385
.
Shabbar
,
A.
, and
M.
Khandekar
,
1996
:
The impact of El Niño–Southern Oscillation on the temperature field over Canada
.
Atmos.–Ocean
,
34
,
401
416
.
Shabbar
,
A.
, and
B.
Bonsal
,
2003
:
An assessment of changes in winter cold and warm spells over Canada
.
Nat. Hazards
,
29
,
173
188
.
Shukla
,
J.
,
1998
:
Predictability in the midst of chaos: A scientific basis for climate forecasting
.
Science
,
282
,
728
731
.
Ting
,
M. F.
,
M. P.
Hoerling
,
T. Y.
Xu
, and
A.
Kumar
,
1996
:
Northern Hemisphere teleconnection patterns during extreme phases of the zonal-mean circulation
.
J. Climate
,
9
,
2614
2633
.
Trenberth
,
K. E.
,
1990
:
Recent observed inter-decadal climate change in the Northern Hemisphere
.
Bull. Amer. Meteor. Soc.
,
71
,
988
993
.
Uppala
,
S. M.
, and
Coauthors
,
2005
:
The ERA-40 Re-Analysis
.
Quart. J. Roy. Meteor. Soc.
,
131
,
2961
3012
,
doi:10.1256/qj.04.176
.
Vincent
,
L. A.
, and
E.
Mekis
,
2006
:
Changes in daily and extreme temperature and precipitation indices for Canada over the twentieth century
.
Atmos.–Ocean
,
44
,
177
193
.
Vincent
,
L. A.
,
X.
Zhang
,
B. R.
Bonsal
, and
W. D.
Hogg
,
2002
:
Homogenization of daily temperatures over Canada
.
J. Climate
,
15
,
1322
1334
.
Wang
,
B.
,
R.
Wu
, and
X.
Fu
,
2000
:
Pacific–east Asia teleconnection: How does ENSO affect east Asian climate?
J. Climate
,
13
,
1517
1536
.
Wang
,
B.
,
Z. W.
Wu
,
J.
Liu
,
C.-P.
Chang
,
J.
Li
, and
T.-J.
Zhou
,
2010
:
Another look at climate variations of the east Asian winter monsoon: Northern and southern modes
.
J. Climate
,
23
,
1495
1512
.
Wu
,
Z. W.
,
J.
Li
,
B.
Wang
, and
X.
Liu
,
2009a
:
Can the Southern Hemisphere annular mode affect China winter monsoon?
J. Geophys. Res.
,
114
,
D11107
,
doi:10.1029/2008JD011501
.
Wu
,
Z. W.
,
B.
Wang
,
J.
Li
, and
F.-F.
Jin
,
2009b
:
An empirical seasonal prediction model of the east Asian summer monsoon using ENSO and NAO
.
J. Geophys. Res.
,
114
,
D18120
,
doi:10.1029/2009JD011733
.
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
,
395
429
.