1. Introduction
Research within various scientific disciplines share a common interest in the study of high-latitude spring onset because of its impact on ecosystem productivity (Linderholm 2006). From a phenological perspective, spring onset, or the onset of the growing season, is identified in terms of local phenomena such as flowering (Asa et al. 2004; D’Odorico et al. 2002) or bug burst (D’Odorico et al. 2002). The phenological onset of the growing season exhibits strong temporal and spatial variability (D’Odorico et al. 2002), which is believed to be related to the cumulative effect of preceding local variations in temperature (White et al. 1997; Schwartz and Crawford 2001). Sparks and Menzel (2002) synthesized data from a wide range of species over a large spatial domain and found that spring onset defined in terms of phenological events (including ground flora, bird migration, tree flowering, and harvest timing) all demonstrate consistent responses to the warming climate. Analyzing data from several stations over western Europe, D’Odorico et al. (2002) attributed the temperature-induced interannual variation of spring onset to the North Atlantic Oscillation (NAO) and concluded that a warmer winter resulting from the positive NAO phase usually leads to an earlier spring onset. Aasa et al. (2004) illustrated a similar relationship for central and eastern Europe; they also related spring onset to different atmospheric circulation patterns. These studies established a robust relationship between phenological spring onset and the large-scale atmospheric circulation.
In addition to influences upon phenology, systematic rapid transitions are also observed to occur in the atmospheric circulation itself. During spring, the Arctic atmosphere undergoes a complicated transition from a statistically steady winter state toward a rapidly evolving warming state that is forced by a combination of increasing solar insolation and internal atmospheric dynamics. During early spring, the Arctic surface air temperature increases dramatically within a short time period. Martin and Munoz (1997) analyzed the Arctic 2-m temperature (T2m) using a polar region–based gridded dataset from the NASA Earth Observing System project Polar Exchange at the Sea Surface (POLES). The daily root-mean-square (RMS) of T2m decreases dramatically during spring, revealing a transition from high wintertime variability to relatively weak springtime variability. This high-frequency behavior is shown in Figs. 1a,b, which display the same analysis calculated from the ECMWF interim reanalysis (ERA-Interim) data. Although the high-frequency intraseasonal variability decreases during spring, the total air temperature field increases quite rapidly as illustrated in the companion Hovmöller plot in Fig. 1c. The rate of T2m increase during spring is much faster than the rate of decrease during fall. In addition, the rapid warming behavior is particularly pronounced in the Arctic, north of 75°N, with less rapid warming observed to the south.
(a) Daily average T2m (°C) over the Arctic from 1979 to 2010. (b) Daily RMS of Arctic T2m (°C). (c) Hovmöller plot of zonal-mean T2m (°C): latitude vs day of the year.
Citation: Journal of Climate 28, 11; 10.1175/JCLI-D-14-00593.1
Studies of the upper-level atmosphere also show an abrupt transition of spring onset. Black et al. (2006) examined stratospheric final warming events and discovered that, during the annual spring demise of the westerly stratospheric polar vortex, the zonal-mean zonal wind decreases rapidly over a vertical domain extending from the midstratosphere downward through the entire troposphere. The near-surface response is dominated by an NAO-like pattern shift from positive to negative phase, with sea level pressure (SLP) rises at polar latitudes. Follow-up diagnostic research revealed that stratospheric final warming events are primarily driven by planetary-scale Rossby waves propagating upward from the troposphere (Black and McDaniel 2007). These studies raised the interesting possibility of a role for the high-latitude stratosphere in enacting the tropospheric circulation changes taking place during spring onset. Further, the studies suggested that dynamical processes likely play a critical role in driving Arctic spring onset within the middle atmosphere. However, our current understanding of the analogous spring onset transition at the surface is limited, at best. Although Martin and Munoz (1997) recognized a rapid increase in surface temperature associated with a decrease in its variability, no previous research efforts have examined the fundamental structures of this rapid seasonal transition. Therefore, the current study is aimed at introducing a metric to isolate the rapid transition of atmospheric Arctic spring onset and characterizing the primary associated circulation structures, laying the foundation for future analyses of the underlying dynamical processes.
Atmospheric science encompasses the internal dynamics of the atmospheric circulation along with interactions among atmospheric systems on various temporal and spatial scales. Spring onset is usually defined as the time when T2m either exceeds a threshold value (Qian et al. 2009, 2011a; Qian et al. 2011b) or undergoes a period of phase change in the seasonal cycle (Thomson 1995). These two approaches provide differing results regarding the long-term trends of spring onset because the underlying physics are different. On the one hand, Qian et al. (2009) applied a temperature threshold approach, which is more closely related to phenological spring onset because of the prominent link between phenological events and absolute surface air temperature, and found evidence for an advancing trend in spring onset linked to global warming and consistent with parallel phenological analyses. They further attributed spring onset variability to annual cycle variations and long-term warming trends (Qian et al. 2011a) and demonstrated an interannual relationship between the time-varying annual cycle and El Niño events (Qian et al. 2011b). On the other hand, Thomson (1995) fitted the seasonal cycle to a predetermined sinusoidal wave function, estimating the phase state from a complex demodulation approach, and found that the phase trend during the last century is consistent with the increasing CO2. However, using the same fitting approach but estimating the phase states via four different methods, Paluš et al. (2005) argued that more evidence is required to link the long-term trend in spring onset to an anthropogenic influence. Synoptic processes such as cyclones and anticyclones are important contributors to sudden temperature increases in high latitudes during spring but it is very difficult to link them directly to phenological events, as pointed out by Aasa et al. (2004). The sinusoidal fitting method, which considers the physical and mathematical features of the time series alone, does not resolve dynamic processes. Therefore, neither approach discussed above is effective in identifying the coherent and rapid large-scale atmospheric transitions from winter to spring: more specifically, the rapid changes in surface air temperature and large-scale circulation. Building on the existing research discussed above, the main goals of the current study are to (i) develop a systematic and robust method to delineate rapid Arctic spring onset (ASO) in the high-latitude atmosphere and (ii) characterize the typical near-surface synoptic behavior occurring during Arctic spring onset. In pursuing (i) we introduce a new approach that effectively captures episodes of Arctic spring onset linked to rapid increases in surface air temperature.
This paper is organized as follows: Section 2 describes the data used and preprocessing procedure. The method used to define rapid Arctic spring onset is introduced in section 3. Section 4 overviews the composite synoptic atmospheric behavior associated with spring onset. Section 5 provides a detailed discussion. Finally, a summary, conclusions, and implications are provided in section 6.
2. Data and general approach
Our study is based on data from ERA-Interim (hereinafter ERAI) (Dee et al. 2011) for 32 yr (1979–2010). To test the sensitivity of our results to the dataset employed, we also identify ASO based on the NASA Modern-Era Retrospective Analysis for Research and Applications (MERRA) data (Rienecker et al. 2011). Unless otherwise mentioned, the results presented are based upon the ERAI data. The daily average T2m is the fundamental field variable used to define ASO. The data from 29 February of every leap year are removed first. Before analyzing the annual temporal behavior, we construct area averages (weighted by the square root of cosine latitude so that the variance at each grid point is weighted by cosine latitude) over the polar cap north of 75°N.
To isolate a rapid warming signal in daily area-averaged T2m for each calendar year, we tested three algorithms: a time derivative method, a radius of curvature method, and a two-phase linear regression. In the first two methods, a 30-day low-pass filter is first applied to the area-averaged daily T2m to remove high-frequency variability while the third method uses unfiltered data. Taking an area average also reduces high-frequency variability, thus helping to overcome the inherent deficiencies of applying 31-day running-mean averages to temporally smooth time series [as noticed by Qian et al. (2011a)]. Further details regarding the three methods are provided in section 3.
The general synoptic characteristics of ASO are studied using composite analyses. Both T2m and T2m anomaly (departure from the climatology) fields are analyzed to study the horizontal warming patterns associated with ASO events. A two-tailed Student’s t test (Wilks 2011) is used to test the statistical significance of anomaly fields. The Mann–Kendall test and Sen’s slope estimator are used to test for the existence of significant long-term trends (Helsel and Hirsch 1992) because of the non-Gaussian distribution of ASO (e.g., see section 4). The smoothed daily T2m climatology is isolated by (i) calculating the long-term climatological-mean T2m for each calendar day and (ii) subsequently applying the 30-day low-pass filter to the resulting climatological cycle. In doing so, we are then able to apply the same methods of isolating rapid warming in the T2m climatological cycle, which will be discussed in section 4. The composite analysis procedure is also later applied to the SLP field in order to characterize the synoptic behavior of ASO.
As will be illustrated in section 4, although the regional warming patterns observed during ASO exhibit considerable interannual variability, they generally fall into three or four pattern categories. As such, we further pursue a synoptic classification and a cluster analysis approaches to group the resulting warming patterns. The synoptic classification is based on a simple visual identification of the location of the primary warming signature. This approach is supplemented by a separate cluster analysis of the warming patterns. To isolate the unique spatial structures of the ASO warming, we specifically use the hierarchical cluster method of Ward (1963), which has been widely used to categorize the Northern Hemisphere wintertime geopotential height patterns (e.g., Cheng and Wallace 1993). The Euclidean distance between the anomaly maps is used as the measure of similarity. At each step, for each cluster, the sum of the Euclidean distance between all maps in the cluster and the centroid of the cluster (which is the average of all the maps within the cluster) is calculated and noted as “the error sum of square” for that cluster. The total of the error sum of square for all clusters is used as a criterion for the merging of next step. All possible pairs of two maps merging together are tested. The pair of maps that merges next is the one that guarantees the minimum increase in the total of the error sum of square for all clusters. Further details of the clustering method can be found in Ward (1963) and Cheng and Wallace (1993).
3. Defining Arctic spring onset
This section describes and compares different algorithms for identifying rapid warming during ASO. We define an ASO index as the time when a rapid increase in T2m occurs over the Arctic. To identify abrupt transition points in each annual cycle of T2m, we explored three methods: the time derivative (a calculus viewpoint), the radius of curvature (a geometric viewpoint), and the two-phase linear regression model (Lund and Reeves 2002) (a statistical viewpoint). The idea behind the first two methods is similar: They identify times marked by an abrupt increase in the rate of change of T2m (which can occur after periods of relatively weak T2m increases or even T2m decreases).
a. The time derivative
From a calculus viewpoint, the “acceleration of T2m” is taken as the second derivative of T2m with respect to time (d2T). Thus, for each year the most rapid warming is taken to be the time when the maximum d2T occurs during a period encompassing late winter through midspring. More specifically, the search for the maximum d2T considers the period from day 55 to 120 of the calendar year. In Figs. 1a,b, the long-term daily average T2m over the Arctic starts to increase while its variability starts to decrease at the end of February. The annual cycle is less fluctuating after late April. Thus, we restrict the search within the period from late February to the end of April, days 55–120. It does not alter the results much if we bring the starting time forward or put the ending time backward.
Whether it is an abrupt transition not only depends on d2T alone but is also influenced by the rate of T2m change (first derivative of T2m with respect to time) at that time. Consider two cases: 1) T2m rate of change increases from 1° to 1.2°C day−1 and 2) T2m rate of change increases from 0.01° to 0.21°C day−1. Although the acceleration of T2m is 0.2(°C day−1) day−1 in both cases, the second case is basically a transition from near-zero T2m change to a dramatic increase. To account for such situations, d2T is further normalized by the first derivative (dT), and d2T/dT is used as a proxy to represent the warming. We take the time of maximum d2T/dT as the ASO index.
b. The radius of curvature
The RoC is positive during periods of T2m acceleration (with respect to calendar day). Thus, when the RoC reaches a minimum positive value, T2m will exhibit the strongest “increase in the T2m increase.” We search for the minimum positive value of RoC during the same period as in the d2T/dT method.
c. The two-phase linear regression
The two-phase linear regression model is most commonly used in identifying changepoints in climate time series (Lund and Reeves 2002). Cook and Buckley (2009) used this method to identify the monsoon season onset, which provides us with a novel perspective for viewing the abrupt ASO transition.
The two methods discussed above consider “temporally localized” information, which only employs T2m during a period of several days around the date of interest and identifies times of acceleration in T2m. The two-phase linear regression approaches the topic by considering the general difference between springtime and wintertime temperature trends. T2m typically increases dramatically during springtime while it remains relatively stable with little increase or decrease during much of the winter. Its variability can be viewed as a high-frequency signal embedded upon a steady mean state in winter and a gradually increasing state in spring. The transition between the two linear states is therefore taken as the ASO in this approach.
To obtain realistic longer-term background states for winter and spring, during each year we retain the unfiltered daily T2m for a relatively long period that begins in December of the year before and continues through June. This period is chosen for the two-phase linear fitting procedure. The identification of changepoints, however, is restricted to the same period as in the first two methods (days 55–120). For each day falling within the identification period, the following algorithm is applied to generate the two-phase linear model:
The specific day is selected as representing the separation between winter and spring periods.
A least squares method is separately applied to the two periods, from 1 December to the separation day and from the separation day to 30 June. The linear fits of the two periods are then taken to approximately represent the background winter and spring states, respectively.
The error associated with this representation is determined by calculating the cumulative sum of the daily root-mean-square difference between the two linear regressions and the T2m series.
Procedure (i)–(iii) is then repeated for each day throughout the identification period. The date when the error estimate reaches its minimum is taken as the most suitable separation between winter and spring or ASO. The two corresponding linear regression fits represent the winter and spring background states of that year.
d. Comparison among the three methods
Figure 2 provides the ASO indices calculated from the three methods for each year in the period of study. The results from the three methods demonstrate considerable similarity. The d2T/dT and RoC methods provide almost identical ASO dates except during 1991 and 2009, which will be discussed below. The correlation coefficient between the two time series is 0.8133 (p = 2.69 × 10−8). Although the fundamental interpretation of ASO is different between the two-phase linear regression model and the other two methods, the results are nonetheless consistent with correlation coefficients of 0.7862 (p = 1.97 × 10−7) and 0.6011 (p = 3.49 × 10−4), respectively, with the RoC and d2T/dT time.
(a) ASO calculated from time derivative (blue), radius of curvature (black), and two-phase linear model (red) methods. (b) ASO during 2009 is presented as an example of the comparison between d2T/dT and RoC methods. T2m (°C) unfiltered (30-day low pass) data are shown with blue (black) solid line; and ASO via RoC (d2T/dT) is shown with a blue asterisk (green square). (c) ASO during 1994 is presented as an example of the comparison between RoC method and the linear regression model. T2m (°C) unfiltered (30-day low pass) data are shown with blue (black) solid line; linear fittings of the winter and spring T2m states from the two-phase linear model are shown with red dashed line; and ASO via the RoC (linear regression model) is shown with a blue asterisk (black square).
Citation: Journal of Climate 28, 11; 10.1175/JCLI-D-14-00593.1
Despite the reassuring consistency of the three methods, we need to select one as the index for our analysis of ASO. Noting the mathematical similarity between (RoC)−1 and d2T/dT, it is no surprise that the two methods provide very similar results. However, during the time of interest, T2m is almost always accelerating and thus d2T is rarely zero; however, there is no guarantee of dT being nonzero, and a near-zero dT will cause problems in calculating d2T/dT. As we find during individual years, it is common for the rapid acceleration of T2m to be embedded within a relatively small background dT. As shown in Fig. 2, the two cases when d2T/dT and RoC give different ASO indices (1991 and 2009 in Fig. 2a) are because of this problem (e.g., 2009 case in Fig. 2b): In this case d2T/dT identifies a time when dT is almost zero resulting in a large d2T/dT, while RoC identifies a time when d2T is truly large. Therefore, considering the inevitable zero dT events, from a practical perspective RoC is more generally applicable than d2T/dT as an optimal method to identify the T2m acceleration times.
Let us now contrast the RoC and two-phase linear regression methods. Figure 2c provides an example of 1994. The ASO identified with the RoC method (blue asterisk) coincides with the local minimum of both the unfiltered (blue solid line) and low-passed filtered (black solid line) T2m data, while the ASO identified with the two-phase linear model (black square) occurs several days earlier during a period of decrease in both unfiltered and low-passed T2m. Selecting among such differences in the dates depends on the scientific problem of interest. The two-phase linear regression is clearly preferable for an investigation of the transition between wintertime and springtime background states, in which the temporally localized T2m behavior matters less than the longer-term trends in T2m. However, since we are interested in identifying periods of relatively rapid T2m increase, the RoC method provides a suitable approach for identifying the initiation of rapid warming periods. Therefore, we choose to adopt the RoC approach in defining ASO.
e. Dataset sensitivity
Before analyzing the characteristics of ASO, the robustness of RoC method is first tested by applying the same algorithm to a second reanalysis dataset, MERRA. Figure 3a contrasts ASO dates derived by applying the RoC method to both ERAI and MERRA datasets. The overall similarity is quite remarkable. For 30 of the 32 events, the two methods provide dates that are within five days of one another. The only two exceptions are 2003 and 2005, which will be discussed further in section 4a.
(a) ASO dates calculated using ERAI (blue) and MERRA (black), respectively. The trend lines of Sen’s slope estimation are plotted in dashed lines of the corresponding color; p values are noted in the corresponding color. (b) Separate early (blue) and late (black) event dates from ERAI. Sen’s trends are shown with a dashed line and p values are in the same color. (c) As in (b), but for MERRA.
Citation: Journal of Climate 28, 11; 10.1175/JCLI-D-14-00593.1
4. Analyzing spring onset
The RoC index defined from section 3 isolates the time each year characterized by a rapid T2m increase, referred to as ASO. This section analyzes fundamental aspects of ASO, including potential long-term trends, the vertical and horizontal structure of associated air temperature change, and circulation changes occurring during ASO.
a. Climatology and trend
The average ASO is day 78 of the calendar year, or 19 March, one day in advance of the vernal equinox. Among the 32 years (1979–2010), the ASO dates exhibit substantial interannual variability, as shown by the black line in Fig. 2a. Onset occurs as early as 1 March and as late as 16 April. The events are not evenly distributed; instead, they appear to be separated into two groups: an early group having a mean onset date at 2 March (day 61 of the calendar year) and a late group with a mean onset date at 3 April (day 93 of the calendar year). This bimodal distribution is reflected via the two concentrated peaks near 6 and 31 March in the histogram plot (Fig. 4a). In addition, in applying the RoC approach to the climatological T2m annual cycle and isolating the first and second RoC minima, we identify two distinct T2m acceleration times in the annual cycle in Fig. 4b. The primary T2m acceleration feature is close to the average date of the late ASO events while the secondary T2m acceleration feature approximately corresponds to the average date of early ASO events. This two-stage onset behavior is also evident in all the individual years considered.
(a) Distribution of ASO dates. (b) Climatology annual cycle of daily Arctic T2m (°C; blue line); primary (secondary) warming signature in the climatological annual cycle indicated by blue square (circle); for comparison, the composite T2m during the ASO is also displayed (red line); the average ASO is shown as a red dot. The most rapid increase of the climatological T2m is identified through the RoC method. In contrast to the identification of rapid T2m increase for a specific year, the analysis of the climatological T2m retains both the minimum and the second minimum RoC signatures and denotes them as the primary and secondary T2m warming signatures, respectively. The primary T2m signature is at day 95 of the calendar year while the secondary is at day 61.
Citation: Journal of Climate 28, 11; 10.1175/JCLI-D-14-00593.1
There is considerable scientific interest in whether the timing of ASO exhibits a statistically significant trend associated with climate change. From a biological viewpoint, studies (e.g., Schwartz et al. 2006; Sparks and Menzel 2002) have ascertained an advancing trend in the ASO. Corresponding studies based upon the atmospheric observational record also indicate a long-term warming trend in Arctic air temperature during spring (Rigor et al. 2000). Consequently, studies defining ASO using a specific temperature threshold also demonstrate an advancing trend during the last century (Qian et al. 2009). As discussed in a number of papers (e.g., White et al. 1997; Schwartz and Crawford 2001), phenological spring onset events result from the preceding cumulative effects of heating; thus, an advancing trend is generally attributed to global warming. However, ASO related to a rapid surface temperature increase represents an abrupt atmospheric seasonal transition, which is distinct from phenological spring onset (even though it may ultimately affect phenology). Thus, there is no a priori expectation that the two definitions will exhibit similar behavior in terms of long-term variability. In addition, because of the bimodal nature of the onset date distribution, we must be very careful in assessing long-term trends in our analysis.
Figure 3a displays the ASO dates and Sen’s trend line fits for both ERAI and MERRA data. Figures 3b,c, on the other hand, show onset dates and trend lines separately for the early and late events for both datasets. Trend values and corresponding p values are summarized in Table 1. In assessing the primary ASO dates (Fig. 3a) over the 32-yr period, the ERAI data exhibit an advancing trend (p = 0.019) while there is no significant trend obtained from the MERRA data. This discrepancy is largely due to two years (2003 and 2005) in which early (late) events are identified in ERAI (MERRA). As shown in Figs. 3b,c and Table 1, when early and late events are assessed separately, no significant trends are identified from either dataset. Therefore, rapid ASO (as determined by the RoC method) does not exhibit recent long-term trends.
The Sen’s trends of ASO (days yr−1) along with the corresponding p values. Trends and p values are calculated for a total of 32 events from 1979 to 2010 as well as separate groups of early and late events from both ERAI and MERRA datasets.
The results of our trend analysis are not necessarily inconsistent with global warming behavior. As global-mean temperature increases, the time when regional temperature exceeds a certain threshold will likely advance. However, as long as the local seasonal cycle in T2m does not change shape appreciably, increasing the mean temperature may shift the cycle upward but does not necessarily alter the typical time when the abrupt transition, or minimum RoC, occurs.
b. Warming structures
The vertical and horizontal structure of the warming pattern occurring during ASO is next analyzed with a composite analysis. For the 32 ASO events, a sequential composite analysis of synoptic variables (e.g., T2m and SLP) is performed over a 45-day period extending from 20 days prior to RoC ASO (denoted as day 0) to 25 days after onset. The timing of ASO varies from case to case but occurs within a period of marked climatological seasonal transition. Field variable anomalies are calculated by first removing the smoothed daily climatology prior to compositing. The resulting composite anomalies include coherent and statistically significant features since the behavior in each case typically differs appreciably from the climatological seasonal trend (annual cycle).
Figure 5 displays the composite time evolution of zonal-mean air temperature anomalies averaged over the latitudinal band extending from 70° to 85°N (weighted by the square root of cosine latitude). The analysis extends from 1000 to 100 hPa. It illustrates the typical vertical structure of the air temperature anomalies during the ASO. We note that the significant temperature anomaly signature is restricted to tropospheric levels and extends no higher than 300 hPa. During the warming period (which begins at day 0), the air temperature anomaly signature is initially negative (below climatological values as in Fig. 4b), weakens in amplitude as temperature approaches climatological values (around day 15), and eventually becomes positive (exceeding climatology) around day 20. The majority of the warming is observed to occur in the 2-week period following onset (day 0).
Composite vertical cross section of zonal-mean air temperature (°C) between 70° and 85°N from 1000 to 100 hPa. Day 0 is the ASO date; negative lags indicate preonset days while positive lags are postonset days. The green line represents the 95% level of confidence in a Student’s t test.
Citation: Journal of Climate 28, 11; 10.1175/JCLI-D-14-00593.1
The “rapidity” of the transition process can be interpreted in two ways: First, the magnitude of the total air temperature increase (climatological seasonal trend plus anomaly) after ASO is greater than that occurring before onset, consistent with the RoC identification procedure. Second, the lower-tropospheric air temperature shifts from a colder-than-climatological state (shown as negative anomalies in Fig. 5 and depicted by the red line in Fig. 4b) to a warmer-than-climatological state (positive anomaly in Fig. 5), indicating a temperature increase that is faster than the climatological-mean seasonal transition in air temperature observed during ASO. This warming occurs mainly in the 15 days after ASO as the temperature anomaly evolves from large negative values to near zero as the total temperature field approaches climatological values.
The composite horizontal structure of the associated surface air temperature changes is shown in Fig. 6, in terms of local changes in both total and anomalous T2m. A remarkably zonally asymmetric structure emerges in both maps. In the left panel we observe that T2m is increasing almost everywhere during ASO (which, in itself, is not surprising). Although T2m increases robustly over the Arctic Ocean, the largest T2m increases occur over the high-latitude portions of the main continents with gradually decreasing magnitudes toward the south. T2m changes over open ocean regions are relatively weak, as might be expected. The right panel displays corresponding changes in the T2m anomaly field after removing the mean seasonal cycle (which is done prior to compositing). Thus, positive values in the right panel represent local T2m increases that exceed climatological T2m increases while negative values indicate regions where T2m increases more gradually than climatology (or even decreases). Similar to the structure in the change of T2m total field, there are significant increases evident over the Arctic and high-latitude continental regions. North of 75°N, an approximate zonally symmetric pattern of strong and statistically significant increases are observed over the Arctic. South of 75°N, considerable zonal asymmetry is observed. In the Western Hemisphere, the T2m increase pattern extends southward to cover portions of the Canadian Northwest Territories and Greenland. In the Eastern Hemisphere, the most prominent signature is a strong T2m increase pattern over northern Siberia. The strong springtime warming observed at high latitudes is consistent with the fact that meteorological variables exhibit larger magnitude seasonal cycles at high latitudes than lower latitudes (Hartmann 1994).
(a) Composite map of T2m change (°C) between days 0 and 15. (b) Composite map of the change in T2m anomaly field (°C) between days 0 and 15; green contour represents the 95% confidence level from a Student’s t test.
Citation: Journal of Climate 28, 11; 10.1175/JCLI-D-14-00593.1
These composite maps isolate the T2m increase pattern common to most the 32 events considered, while case-to-case variability is removed. As will be discussed in the following subsection, the individual warming patterns demonstrate remarkable case-to-case (interannual) variability. Nonetheless, almost all the events exhibit the robust T2m increase observed in the composite analysis over northern Siberia. Therefore, we hereafter denote the region within 60°–90°N, 30°–120°E as the “critical” warming region. Generally speaking, northern Siberia is characterized by marked surface variability. Past studies have concluded that the semipermanent system over the critical region of northern Siberia, the Siberian high, is the most important large-scale circulation system occurring during boreal winter and strongly controls temperature and precipitation variability on multiple time scales (Gong and Ho 2002, 2004).
c. Categorizing events
1) Synoptic classification
As is shown in Fig. 6, the strongest warming in the composite analysis occurs over the critical region. However, a study of individual cases reveals that there are distinct regional flavors in terms of the primary large-scale warming regions. A synoptically based algorithm is applied to each T2m anomaly increase map in order to assess the primary warming regions: For each event, contiguous regions of large T2m increases spanning at least 45° longitude and 12° latitude are first identified. We then categorize events into groups according to the regional central location(s) of the contiguous regions identified. We find that warming signatures commonly occur in four high-latitude (60°–90°N) sectors: the critical region over northern Siberia, the Greenland–North American region, East Asia, and Alaska. We note that multiple warming regions are sometimes active during an individual event. In such cases, we assign that event to each category in which it fits. Thus, we end up with 17 cases that demonstrate warming signals over the critical region, 11 over Greenland–North America, 10 over East Asia, and 7 over Alaska. As discussed above, the four categories are not mutually exclusive. For example, since T2m anomaly warming occurs over both the critical region and Alaska during spring 1979, we assign the 1979 event to both categories in our analyses.
Figure 7 provides maps of the composite 15-day change in T2m anomaly for the four groups. In Fig. 7a, the composite map for the 17 critical region events demonstrates the strong warming signal covering much of western Siberia. Compared with the composite map of all 32 events, more than ⅔ of the critical region is controlled by the robust temperature increase (at the 95% confidence level) while the Western Hemisphere has a much weaker warming signal in these cases. Figure 7b shows the composite map for the Greenland–North America events. The warming pattern over the Western Hemisphere extends much farther south into North America and eastward to the western edge of the Greenland plateau. A relatively weak but significant warming signature is found over the eastern portion of the critical warming region. The composite map for the East Asian events is provided in Fig. 7c. T2m anomaly increases are observed over portions of the Arctic Ocean, East Siberian Sea, and high-latitude Eurasian continent. Two substantial warming centers are found over East Asia and a portion of the critical region, respectively. Figure 7d displays the analogous composite map for the Alaskan category of events. These cases exhibit a broad statistically significant warming pattern that extends over the Bering Sea, Chukchi Sea, northern Alaska, and a large portion of the Arctic Ocean. A significant warming signature is also observed in the southern portion of the critical region. To summarize, besides the critical region cases (Fig. 7a), each of the other three groups have warming patterns that are characterized by (i) a primary warming center over the region indicated by the group name and (ii) a secondary center that overlaps the critical warming region identified in the composite analysis of all cases (Fig. 6b). The differences noted in the primary warming features for each category suggest important interannual differences in the factors leading to rapid T2m increases while the shared warming feature over the critical region suggests that a common forcing mechanism exists in this region.
As in Fig. 6b, but for the four groups from synoptic classification: (a) critical region, (b) Greenland–North America, (c) East Asia, and (d) Alaska.
Citation: Journal of Climate 28, 11; 10.1175/JCLI-D-14-00593.1
2) Cluster analysis
To more objectively categorize events, a hierarchical cluster analysis using Ward’s method is applied to the 32 annual maps of 15-day change in T2m anomaly during ASO. Figure 8 displays the last three merging steps of the cluster analysis. Two primary cluster branches of similar sample size (15 and 17) emerge at the last step in the analysis. The cluster having 15 events consists of three subbranches, of which two are very small in size (2 and 3 events, respectively) while the other is relatively large (10 events). The four clusters deemed of primary interest are the critical region (CR), the North American continent (NAmerica), the North America critical region (NAmerica-CR), and upstream of critical region (up-CR). These designations are based on the location of main T2m anomaly warming signatures in the corresponding composite maps (Fig. 9).
Cluster tree of the last three steps in the cluster analysis.
Citation: Journal of Climate 28, 11; 10.1175/JCLI-D-14-00593.1
As in Fig. 7, but for the four groups obtained from the cluster analysis: (a) CR, (b) NAmerica, (c) up-CR, and (d) NAmerica-CR. Since the sample sizes of up-CR and NAmerica-CR categories are small, a statistical test is not performed for these two categories and discussion is mainly focused on the other two categories.
Citation: Journal of Climate 28, 11; 10.1175/JCLI-D-14-00593.1
Despite the small sample size, the NAmerica-CR cluster (two elements) and up-CR cluster (three elements) each consists of events closely resembling one another. The NAmerica cluster exhibits a robust warming signature over the North American sector extending through the northern edge of Greenland toward the North Pole. This pattern has a strong overlap with that associated with the Greenland–North America category in our earlier results of synoptic classification (Fig. 7b). More than half of the 32 events lie in the CR cluster, which is characterized by a prominent warming signature over the critical region as found in the 32 case composite map (Fig. 6). In the CR composite map (Fig. 9a), the largest-amplitude warming signature occurs over the critical region with a weaker warming signal downstream north of the Bering Strait and Alaska.
In categorizing the events from synoptic classification, we ended up with one relatively small category characterized by warming in the Alaska sector. No single group having such character appears as a unique branch in our more objective cluster analysis. However, the composite map for the CR cluster does, in fact, have a warming signal near the Alaska sector. Considering earlier merging steps in our cluster tree, we find that five events with warming over Alaska are clustered together in the first few steps of the cluster analysis and later on merged with other clusters into the CR cluster. That is to say, events having Alaska warming 1) share a very similar spatial pattern that is isolated in the early steps of our cluster analysis and 2) also contain a nonnegligible warming signature in the critical region (as confirmed by examining individual maps or the Alaska composite in Fig. 7d) so that they are eventually folded into the CR cluster during the later steps of our cluster analysis. Four events with warming over East Asia exhibit behavior similar to the Alaska warming events and they are also merged into the CR cluster during the early steps. The events characterized by Alaska and East Asia warming are indicated in the cluster tree in gray even though they are not separate categories in the final clusters.
The reader should keep in mind that in the cluster analysis events are merged together because of their similarity in spatial structure and each event is only allowed to belong to one cluster. However, the synoptic classification allows individual events to be placed in more than one category. Therefore, the fact that several of the 17 events in CR cluster have warming signatures over Alaska (5 events) or East Asia (4 events) does not imply that critical region warming is restricted to only the other 8 events in the CR cluster. Instead, a strong warming signature over the critical region is common among all CR events. Generally speaking, although the results from the synoptic classification and objective categorization approaches do not precisely correspond, the objective cluster analysis does similarly isolate the dominant regions of the rapid T2m increase during ASO. It also suggests a complex case-to-case variability in ASO. Since the synoptic classification not only allows for a larger sample size within each group but also provides a well-defined spatial separation among the primary warming regions, our following composite analysis is focused on the synoptically categorized events.
d. Evolution of circulation
We next study how the large-scale atmospheric circulation typically evolves during the ASO. The semipermanent pressure systems—namely, the Aleutian low, the Siberian high, and the Icelandic low—dominate atmospheric variability during boreal winter. In a climatological sense, the magnitude of these surface pressure systems weakens during spring (e.g., Fig. 10d), resulting in large-scale changes in the atmospheric circulation. We first study the composite behavior considering all 32 annual events as a single group.
(a) Composite map of SLP (hPa) at days (a) −5 and (b) 15. (c) Composite map of the SLP difference (hPa) between days −5 and 15. (d) Composite map of difference in the SLP climatology (hPa) between days −5 and 15.
Citation: Journal of Climate 28, 11; 10.1175/JCLI-D-14-00593.1
Prior to ASO (Fig. 10a), both the Aleutian low and the Siberian high are stronger in magnitude compared to climatology (not shown) while the Icelandic low is weaker and shifted eastward compared to climatology. During ASO, both semipermanent low systems shift eastward in relation to the climatological features. By day +15 (Fig. 10b), the center of Icelandic low moves northeastward, extending into the Kara Sea. Figure 10c displays the map of the composite SLP difference between days +15 and −5. The change in the Icelandic low is reflected by the SLP increase over the Atlantic–western Europe coast and the broader area of SLP decrease extending through the Barents Sea, Kara Sea, and northernmost Eurasia. Meanwhile, south of the critical region, the Siberian high markedly weakens and shifts northeastward, in association with a significant SLP decrease over central Siberia along with a region of little or no SLP change to the northeast of the initial high pressure center. Interestingly, another region of significant SLP decrease is observed over the midlatitude East Asian coast. This is a manifested by a weakened Siberian high and a westward extension of the tail of the Aleutian low. The central pressure of the Aleutian low increases quite markedly; however, the resulting low pressure pattern extends zonally upstream through the western Pacific into East Asia, in association with significant SLP increases over the central North Pacific along with pronounced SLP decreases over easternmost Asia.
The anomalous advection is mainly due to the interaction of the anomalous horizontal circulation with the climatological-mean temperature field. Figure 11 superposes composite wind vector anomalies upon the climatological-mean T2m field averaged over the 15-day period following ASO (when maximum warming occurs). A systematic southwesterly flow is observed in the southern critical region while southeasterly is found within the eastern portion of the critical region. Meanwhile, relatively weak southeasterly wind anomalies are observed over the coastal region of East Asia. The cross gradient anomalous circulation results in robust warm advection over a large portion of the critical region.
Composite of anomalous wind vectors at 925 hPa (m s−1) and T2m climatology (K) averaged between days 0 and +15.
Citation: Journal of Climate 28, 11; 10.1175/JCLI-D-14-00593.1
During the period of ASO, the Icelandic low and Siberian high jointly provide a northward pathway within the critical region for warm advection in association with southerly geostrophic flow. Meanwhile, the emerging low pressure system along coastal East Asia helps to bring relatively warm and moist air onshore from the western North Pacific. These two pathways of warm advection appear to be the primary source of rapid warming over the critical region. In addition, the concurrent weakening Siberian high results in less favorable conditions for cold air mass formation thereby reducing the inland cold air source. Our results provide an initial synoptic perspective of the rapid ASO warming and suggest an important role of dynamical processes in ASO.
The same synoptic analysis is additionally applied to each of the four categories from our synoptic classification scheme (Fig. 12). The evolving behavior of the surface semipermanent synoptic features is qualitatively similar to the composite results discussed above for all 32 years (Fig. 10): The intensifying southerly flow located east of the migrating Icelandic low along with the easterly flow associated with deepening low pressure over the westernmost North Pacific provide warm advection, and the weakening Siberian high reduces the potential for cold air mass formation. As discussed in section 4c, the critical region serves as a secondary warming region in the other three ASO categories. Parallel behavior in the Siberian high, Icelandic low, and East Asia coastal low pressure is also observed in the other categories, suggesting a common synoptic condition that favors northern Siberia as a critical region for the rapid increase in surface temperature during ASO.
As in Figs. 10a–c, but for the four groups from synoptic classification: (a1)–(a3) critical region, (b1)–(b3) Greenland–North America, (c1)–(c3) East Asia, and (d1)–(d3) Alaska.
Citation: Journal of Climate 28, 11; 10.1175/JCLI-D-14-00593.1
Besides the critical region, the primary warming region in each of the other three categories exhibits unique synoptic evolution in the SLP field. During ASO for the Greenland–North America category, when warming is concentrated over northern Canada into westernmost Greenland, high pressure over North America and the pole weakens and shrinks in areal extent while low pressure spreads into the continental margins from the east and west. Southwesterly flow associated with the eastward migrating Aleutian low serves to transport warm moist air from the Gulf of Alaska into the continental interior. At the same time, easterly flow along the northern branch of Icelandic low carries warm moist air across Greenland into Baffin Bay, warming adiabatically upon descent. The two warm advection pathways feed into the primary warming region, with a concomitant weakening of continental high pressure and cold air mass formation. The SLP evolution of the Greenland cases (Figs. 12b1–b3) resembles that of the critical region category (Figs. 12a1–a3). This similarity carries over to the T2m anomaly change patterns: Comparing Figs. 7a,b, both categories demonstrate robust T2m increases in the critical region and the Greenland–North American sector. While the relative magnitudes of these two warming features differ, they vary in direct relation to the concomitant differences in the regional amplitude of the SLP changes occurring in the two warming regions.
The other two categories exhibit very different synoptic behavior compared with the critical region and Greenland–North American groups. For East Asian events, the Siberian high weakens but extends northeastward to the east coast of Russia resulting in a significant SLP increase along the coast. This high pressure extension not only increases the SLP locally over East Asia but also prevents the low pressure disturbance from extending into the eastern continental margin. Thus, the oceanic low pressure system is restricted to the western Pacific and the associated circulation pathway leads to the transport of warm moist air across the Bering Sea and the Sea of Okhotsk into East Asia. In the Alaskan composite, the Aleutian low shifts slightly westward while the surface high pressure over the North American continent strengthens and extends into the westernmost North Pacific (a behavior opposite to that observed in the critical region composite). In this case, the associated geostrophic flow between the Aleutian low and continental effectively transports warm air from the Gulf of Alaska.
To summarize, our synoptic analyses indicate that the rapid regional warming signatures observed during the different types of ASO are closely tied to atmospheric circulation changes brought about by relatively rapid alterations in the strength and structure of well-recognized semipermanent synoptic features, including the Aleutian low, Icelandic low, and Siberian high. Heat transport resulting from these synoptic patterns, particularly the advection of climatological temperature by the anomalous circulation field, plays a prominent role in transporting heat to the Arctic latitudes during rapid spring onset.
5. Discussion
Contrary to the general expectation of an advancing trend in spring onset associated with a warming background climate, we find strong interannual variability in spring onset but no evidence of significant long-term trends. Wiltshire and Manly (2004) and Wiltshire et al. (2008) analyzed spring onset in terms of phytoplankton bloom and also ascertained no significant trend with a comparable interannual variability. A recent study by Lohmann and Wiltshire (2012) linked the early and late spring onsets with the two major types of large-scale atmospheric regimes over the high latitudes of the Atlantic–European sector. This reinforces the notion that the robust interannual variability of spring onset timing suggests a key role for atmospheric dynamical processes in determining the seasonal transition time in individual years. Somewhat unexpectedly, the spring onset dates exhibit a bimodal distribution (Fig. 3a). As is discussed in section 3, the bimodal distribution parallels the two-stage warming behavior that is inherent in the T2m climatological cycle, itself. A study of individual years reveals that this two-stage process is very common. Our RoC algorithm differentiates between the two by selecting the stage with the greatest T2m acceleration. In a climatological sense, the later-stage onset dominates; however, during individual years, it can be either of the two stages, depending upon the unique T2m evolution for each year. We note that the parallel results obtained from the two-phase linear regression method provide a similar bimodal representation of interannual spring onset behavior. A closer study of individual two-stage events reveals that, in years characterized by early (late) spring onset dates, the later (earlier) stage appears as a perturbation embedded in the springtime warming (steady wintertime) state. One interesting finding is that the average spring onset date, 19 March, is only one day away from vernal equinox, suggesting a prominent general role for the climatological annual solar cycle in determining the period of spring onset.
Using the index denoting ASO dates, we then studied the spatial structure of T2m changes and associated SLP evolutions for the first time. To isolate common recurrent patterns, we used two different methods to categorize warming events into groups having similar spatial structures: The first method identifies contiguous horizontal regions of T2m increase and then classifies events in terms of the central location of the warming regions. The second method applies a hierarchical cluster analysis based on Ward’s method, objectively grouping similar events into the same category. Although the results obtained from the two methods differ, there is a remarkable similarity in the overall categorization. The two main branches in the cluster analysis correspond very well with the critical region category and the Greenland–North America category, respectively, obtained in the synoptic classification. The Alaska and East Asia warming categories obtained in the first classification method lie in the CR group of the cluster analysis. The remaining discrepancies mainly come from the different foundations of the two methods.
Despite slight differences in the results obtained via the two methods, both approaches uncover previously unrecognized regional warming structures associated with ASO, which can be categorized into four distinct groups. The unique spatial structure observed in each category is additionally linked to regional differences in the circulation anomaly patterns. In the Arctic, the springtime atmospheric circulation is strongly influenced by stratospheric final warming events, when the stratospheric polar vortex breaks down in association with rapid warming at stratospheric altitudes (Andrews et al. 1987). During final warming events, the stratospheric circulation changes extend downward to tropospheric altitudes, leading to significant regional changes in the large-scale tropospheric circulation (Black et al. 2006). Therefore, we originally hypothesized that Arctic spring onset may be linked to stratospheric final warming events. Although coherent decreases in zonal wind throughout the troposphere and stratosphere are observed during Arctic spring onset (not shown), we find no evidence for a causal link between stratospheric final warming events and rapid surface warming associated with Arctic spring onset. The observed rapid increase in air temperature is confined to below the tropopause with a robust signal extending from the surface to the middle troposphere.
To further explore the tropospheric synoptic behavior of Arctic spring onset, we perform composite analyses of SLP. While the semipermanent surface pressure systems demonstrate a weakening trend as the climatological circulation transitions from a steady-state cold season toward the warm season, the anomalous circulation observed during spring onset events does not simply represent an amplification of the climatological trend. Instead, unique regional behavior in the SLP anomaly field provides favorable synoptic conditions for regional rapid warming to occur. For the critical region and Greenland–North America events, the dominant wintertime continental high pressure system weakens dramatically while the adjacent oceanic low pressure system spreads inland resulting in a new large-scale state characterized by (i) weakened cold air mass formation and (ii) increased warm advection, which both contribute to rapid continental warming. For the Alaskan and East Asian events, both the coastal high pressure system and oceanic low pressure system strengthen leading to a stronger geostrophic onshore flow. One complication in understanding the circulation change that takes place during spring onset events is the role of the background climatological seasonal trend (i.e., the climatological seasonal cycle). The SLP anomaly change studied here is embedded within a gradually evolving background state. However, composite changes in the total (climatology plus anomaly) SLP field observed during spring onset exhibit a general similarity with parallel regional changes in the SLP anomaly field. This similarity suggests that climatological seasonal transition only provides a minor contribution to the abrupt spring onset events studied here.
The strong coupling between the evolution of the SLP and T2M fields is consistent with such coupled behavior observed in earlier studies. For example, studies of the Siberian high reveal that cold surface conditions are essential to an intense high Siberian high (Takaya and Nakamura 2005) while diabatic cooling occurring within the high pressure system provides a source of surface cold air (Ding and Krishnamurti 1987; Ding 1990). Our analysis of the rapid surface warming represents a decreasingly favorable situation for the intensification of the Siberian high and formation of a cold polar air mass. Similarly, the study of anticyclogenesis over the Alaska region by Bodurtha (1952) identifies a preexisting surface cold air mass subsequently altered by coherent warm advection from the west, which is consistent with the synoptic conditions we observe for our Alaska warming events.
The temperature increase can be further analyzed in terms of the thermodynamic equation in which the local temperature tendency can be decomposed into linear advective, nonlinear advective, adiabatic, and diabatic components. Our synoptic analysis of the anomalous surface circulation in relation to the climatological-mean surface temperature field (Fig. 11) provides initial evidence that linear advection of climatological-mean temperature by the anomalous circulation likely provides an important contribution to the local rapid warming that occurs in southern and eastern portions of the critical region. In a portion of the northern critical region (from the Laptev Sea westward covering most of the Kara Sea and parts of the Barents Sea) the anomalous easterly circulation results in cold advection that locally opposes warming. However, nonlinear contributions linked to enhanced baroclinic eddy activity may also play a role.
On interannual or decadal time scales the springtime atmospheric circulation is influenced by winter low-frequency modes such as the northern annular mode (Liu and Ding 2007; Stine and Huybers 2012) and the Pacific–North American pattern (Stine and Huybers 2012). However, during Arctic spring onset, we find no statistically significant change in any of the leading teleconnection indices, even though there is a strong and significant decrease in SLP covering most Arctic and the critical region (Fig. 10c). In midlatitudes, there are significant SLP anomaly increases over the eastern North Atlantic and North Pacific. Although these structures bear some resemblance to the NAO pattern, the midlatitude centers are longitudinally phase shifted compared to the NAO. The results suggest a complex dynamics that connects Arctic spring onset to distinct large-scale circulation anomaly patterns at synoptic and intraseasonal time scales.
6. Conclusions
In summary, we have developed a robust method for identifying Arctic spring onset (ASO) events in terms of surface air temperature evolution and have characterized the basic structure and synoptic evolution of Arctic spring onset. To identify the rapid temperature transition serving as ASO, we applied methods from two perspectives. First, we consider the abrupt transition to be a T2m acceleration that is isolated in terms of the local radius of curvature (RoC). Second, we consider ASO as a T2m transition from a wintertime steady state to a warming springtime state using a two-phase linear regression model to isolate the two background states. Although each approach has its pros and cons, the ASO dates obtained from the two different methods are highly correlated. Since our analyses concentrate upon surface temperature increase during ASO, we deem the first method to be most suitable and mainly employ the RoC approach in our study.
The rapid warming observed over polar latitudes during ASO is roughly zonally symmetric while the warming patterns observed farther south show more regional localization with northern Siberia identified as a critical region in which virtually all cases demonstrate a significant warming signature during ASO. However, regional warming patterns observed outside of the critical region during ASO vary noticeably from year to year, and thus the 32 events are additionally categorized based on the primary warming regions. The synoptic evolution of sea level pressure for the composite of all 32 events as well as the four separate categories demonstrates a strong coupling between the SLP and T2m fields, which strongly suggests that ASO events are dynamically driven by large-scale atmospheric processes.
Although our analyses implicate atmospheric dynamical processes as important players in Arctic spring onset, additional diagnostic research is required in order to ascertain the precise physical nature of such events. To this end, our future research efforts will focus on pursuing dynamical diagnostic studies in order to obtain a comprehensive understanding of the physical mechanisms that are responsible. This will include a detailed heat budget analysis in which we will quantitatively assess advective (linear and nonlinear), diabatic, and adiabatic contributions to the regional lower-tropospheric temperature changes occurring during Arctic spring onset.
Acknowledgments
We are grateful to Walter A. Robinson and Xiangdong Zhang for their discussion on this topic. We also thank the three anonymous reviewers for their insightful comments and suggestions on this paper. This material is based upon work supported by the National Science Foundation under Grant ARC-1107384.
REFERENCES
Aasa, A., J. Jaagus, R. Ahas, and M. Sepp, 2004: The influence of atmospheric circulation on plant phenological phases in central and eastern Europe. Int. J. Climatol., 24, 1551–1564, doi:10.1002/joc.1066.
Andrews, D. G., J. R. Holton, and C. B. Leovy, 1987: Middle Atmosphere Dynamics. Academic Press, 489 pp.
Black, R. X., and B. A. McDaniel, 2007: The dynamics of Northern Hemisphere stratospheric final warming events. J. Atmos. Sci., 64, 2932–2946, doi:10.1175/JAS3981.1.
Black, R. X., B. A. McDaniel, and W. A. Robinson, 2006: Stratosphere–troposphere coupling during spring onset. J. Climate, 19, 4891–4901, doi:10.1175/JCLI3907.1.
Bodurtha, F. T., 1952: An investigation of anticyclogenesis in Alaska. J. Meteor., 9, 118–125, doi:10.1175/1520-0469(1952)009<0118:AIOAIA>2.0.CO;2.
Cheng, X. H., and J. M. Wallace, 1993: Cluster analysis of the Northern Hemisphere wintertime 500-hPa height field: Spatial patterns. J. Atmos. Sci., 50, 2674–2696, doi:10.1175/1520-0469(1993)050<2674:CAOTNH>2.0.CO;2.
Cook, B. I., and B. M. Buckley, 2009: Objective determination of monsoon season onset, withdrawal, and length. J. Geophys. Res., 114, D23109, doi:10.1029/2009JD012795.
Dee, D. P., and Coauthors, 2011: The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quart. J. Roy. Meteor. Soc., 137, 553–597, doi:10.1002/qj.828.
Ding, Y., 1990: Build-up, air mass transformation and propagation of Siberian high and its relations to cold surge in East Asia. Meteor. Atmos. Phys., 44, 281–292, doi:10.1007/BF01026822.
Ding, Y., and T. N. Krishnamurti, 1987: Heat-budget of the Siberian high and the winter monsoon. Mon. Wea. Rev., 115, 2428–2449, doi:10.1175/1520-0493(1987)115<2428:HBOTSH>2.0.CO;2.
D’Odorico, P., J. Yoo, and S. Jaeger, 2002: Changing seasons: An effect of the North Atlantic Oscillation? J. Climate, 15, 435–445, doi:10.1175/1520-0442(2002)015<0435:CSAEOT>2.0.CO;2.
Gong, D. Y., and C. H. Ho, 2002: The Siberian high and climate change over middle to high latitude Asia. Theor. Appl. Climatol., 72, 1–9, doi:10.1007/s007040200008.
Gong, D. Y., and C. H. Ho, 2004: Intra-seasonal variability of wintertime temperature over East Asia. Int. J. Climatol., 24, 131–144, doi:10.1002/joc.1006.
Hartmann, D. L., 1994: Global Physical Climatology. Academic Press, 411 pp.
Helsel, D. R., and R. M. Hirsch, 1992: Statistical Methods in Water Resources. Elsevier, 522 pp.
Linderholm, H. W., 2006: Growing season changes in the last century. Agric. For. Meteor., 137, 1–14, doi:10.1016/j.agrformet.2006.03.006.
Liu, X. H., and R. Q. Ding, 2007: The relationship between the spring Asian atmospheric circulation and the previous winter Northern Hemisphere annular mode. Theor. Appl. Climatol., 88, 71–81, doi:10.1007/s00704-006-0231-y.
Lohmann, G., and K. H. Wiltshire, 2012: Winter atmospheric circulation signature for the timing of the spring bloom of diatoms in the North Sea. Mar. Biol., 159, 2573–2581, doi:10.1007/s00227-012-1993-7.
Lund, R., and J. Reeves, 2002: Detection of undocumented changepoints: A revision of the two-phase regression model. J. Climate, 15, 2547–2554, doi:10.1175/1520-0442(2002)015<2547:DOUCAR>2.0.CO;2.
Martin, S., and E. A. Munoz, 1997: Properties of the Arctic 2-meter air temperature field for 1979 to the present derived from a new gridded dataset. J. Climate, 10, 1428–1440, doi:10.1175/1520-0442(1997)010<1428:POTAMA>2.0.CO;2.
Paluš, M., D. Novotná, and P. Tichavský, 2005: Shifts of seasons at the European mid-latitudes: Natural fluctuations correlated with the North Atlantic Oscillation. Geophys. Res. Lett., 32, L12805, doi:10.1029/2005GL022838.
Qian, C., C. Fu, Z. Wu, and Z. Yan, 2009: On the secular change of spring onset at Stockholm. Geophys. Res. Lett., 36, L12706, doi:10.1029/2009GL038617.
Qian, C., C. Fu, Z. Wu, and Z. Yan, 2011a: The role of changes in the annual cycle in earlier onset of climatic spring in northern China. Adv. Atmos. Sci., 28, 284–296, doi:10.1007/s00376-010-9221-1.
Qian, C., Z. Wu, C. Fu, and D. Wang, 2011b: On changing El Niño: A view from time-varying annual cycle, interannual variability, and mean state. J. Climate, 24, 6486–6500, doi:10.1175/JCLI-D-10-05012.1.
Rienecker, M. M., and Coauthors, 2011: MERRA: NASA's Modern-Era Retrospective Analysis for Research and Applications. J. Climate, 24, 3624–3648, doi:10.1175/JCLI-D-11-00015.1.
Rigor, I. G., R. L. Colony, and S. Martin, 2000: Variations in surface air temperature observations in the Arctic, 1979–97. J. Climate, 13, 896–914, doi:10.1175/1520-0442(2000)013<0896:VISATO>2.0.CO;2.
Schwartz, M. D., and T. M. Crawford, 2001: Detecting energy-balance modifications at the onset of spring. Phys. Geogr., 22, 394–409.
Schwartz, M. D., R. Ahas, and A. Aasa, 2006: Onset of spring starting earlier across the Northern Hemisphere. Global Change Biol., 12, 343–351, doi:10.1111/j.1365-2486.2005.01097.x.
Sparks, T. H., and A. Menzel, 2002: Observed changes in seasons: An overview. Int. J. Climatol., 22, 1715–1725, doi:10.1002/joc.821.
Stine, A. R., and P. Huybers, 2012: Changes in the seasonal cycle of temperature and atmospheric circulation. J. Climate, 25, 7362–7380, doi:10.1175/JCLI-D-11-00470.1.
Takaya, K., and H. Nakamura, 2005: Mechanisms of intraseasonal amplification of the cold Siberian high. J. Atmos. Sci., 62, 4423–4440, doi:10.1175/JAS3629.1.
Thomson, D. J., 1995: The seasons, global temperature, and precession. Science, 268, 59–68, doi:10.1126/science.268.5207.59.
Ward, J. H., 1963: Hierarchical grouping to optimize an objective function. J. Amer. Stat. Assoc., 58, 236–244, doi:10.1080/01621459.1963.10500845.
White, M. A., P. E. Thornton, and S. W. Running, 1997: A continental phenology model for monitoring vegetation responses to interannual climate variability. Global Biogeochem. Cycles, 11, 217–234, doi:10.1029/97GB00330.
Wilks, D. S., 2011: Statistical Methods in the Atmospheric Sciences. 3rd ed. Elsevier/Academic Press, 676 pp.
Wiltshire, K. H., and B. F. J. Manly, 2004: The warming trend at Helgoland Roads, North Sea: Phytoplankton response. Helgol. Mar. Res., 58, 269–273, doi:10.1007/s10152-004-0196-0.
Wiltshire, K. H., A. M. Malzahn, K. Wirtz, W. Greve, S. Janisch, P. Mangelsdorf, B. F. J. Manly, and M. Boersma, 2008: Resilience of North Sea phytoplankton spring bloom dynamics: An analysis of long-term data at Helgoland Roads. Limnol. Oceanogr., 53, 1294–1302, doi:10.4319/lo.2008.53.4.1294.
