Impacts of the Pacific decadal oscillation (PDO) on North American climate were initially assessed over one negative (~1943 to 1976) and one positive (1977 to ~1990) PDO regime. Release of the Twentieth Century Reanalysis and the recent occurrence of negative PDO years make it possible to study the stability of PDO teleconnections. This analysis identified consistency in broad-scale teleconnection patterns but also critical differences in the amplitude of circulation pattern, temperature, and precipitation anomalies between comparable phases of the PDO. Many of these discrepancies were apparent after controlling for long-term trends and the impact of ENSO and were associated with variability in Atlantic Ocean temperatures and in the northern annular mode. Results from this study suggest that not all of the climate variability attributed to the PDO derives solely from fluctuations in Pacific sea surface temperatures (SSTs), that the climatic impact of these SST anomalies varies over time, or that the PDO might have a “mixed” state with muted teleconnections. Any of these conclusions has substantial implications for reconstruction of the PDO and its use to understand past hydrologic or ecological changes. They suggest that evaluation of climate models on the basis of their ability to simulate teleconnection patterns of low-frequency modes of climate variability should be undertaken with the recognition that observational records may not be long enough to capture the full range of variability in teleconnection patterns.
In North America, decadal-scale variability in climate is commonly associated with coupled ocean–atmosphere variability in the North Pacific (e.g., Mantua and Hare 2002; Gershunov and Barnett 1998). Regional fluctuations in sea surface temperature and sea level pressure patterns are described by a number of indices and often referred to as Pacific decadal variability (PDV). The Pacific decadal oscillation (PDO) is one metric of PDV, and it is probably the most commonly used PDV index outside of climatology. Although there are differences of opinion about the driver or drivers of the PDO (Strong and Magnusdottir 2009; Alexander 2010), there is a large body of work documenting the impact of the PDO on North American climate, hydrology, and ecology.
The PDO is typically defined as the first empirical orthogonal function (EOF) of sea surface temperature (SST) in the Pacific basin north 20°N. Calculations are performed after adjusting Pacific Ocean SST for trends in global SST (Mantua and Hare 2002). The negative or “cold” phase of the PDO is characterized by cooler than normal SSTs along the west coast of North America and throughout the tropical Pacific, warmer temperatures in the central and western North Pacific, and a weakened Aleutian low. The reverse characterizes “warm” or positive PDO conditions (Mantua et al. 1997; Mantua and Hare 2002).
There is extensive literature discussing the drivers of the PDO. Several studies have identified links between the PDO and the tropical Pacific (e.g., Newman et al. 2003; Deser et al. 2004), while others suggest that feedbacks between the atmosphere and ocean in the midlatitudes are critical (e.g., Latif and Barnett 1994; Schneider et al. 2002). In a recent comprehensive review, Alexander (2010, p. 123) indicated that multiple processes are important in creating the variability recognized as the PDO, concluding that “unlike ENSO, the PDO does not appear to be a mode of the climate system, but rather it results from several different mechanisms.”
Irrespective of their primary drivers, the examination of which is outside the scope of this study, the SST and circulation anomalies attributed to the PDO appear to influence climate over North America. In an analysis spanning 1950 to 1996, Mantua and Hare (2002) found positive correlations between the PDO and November to March temperatures across much of the western contiguous United States and negative correlations with temperature in the eastern United States, particularly in the southeast. Papineau (2001) found cooler winters in Alaska during negative PDO years and, similar to findings for the western contiguous United States (e.g., Wise 2010; Gershunov and Barnett 1998), also found that the phase of the PDO altered the expression of La Niña and El Niño events in Alaska. These findings were corroborated by Hartmann and Wendler (2005) and Fleming and Whitfield (2010), who found that the warm phase of the PDO was associated with warmer temperatures in Alaska and Canada. Analysis by Duffy et al. (2005) also suggested that the positive PDO was associated with warmer summer conditions in interior Alaska. Fleming and Whitfield (2010) found that wetter conditions tended to occur during positive PDO regimes in the winter and/or spring in southern Alaska and parts of coastal British Columbia. Many of these relationships were recently confirmed by Mills and Walsh (2013), who also highlighted the potential for intraseasonal variability in climatic anomalies associated with the PDO.
These studies, although critically important for understanding North America’s climate and instigating the analysis of low-frequency variability in the climate system, share a significant and, until recently, largely unavoidable limitation. The vast majority of the climate data used to describe the PDO’s teleconnections date from no earlier than the 1950s or late 1940s. In many parts of the United States and Canada, weather and climate data are less readily available prior to the mid-twentieth century, and few reanalysis products cover the first half of the last century. Although the existing research identifies strong contrasts between negative and positive phases of the PDO, most of the studies are effectively sampling one negative or cool PDO regime (about 1942 to 1976) and one positive or warm PDO phase (1977 to about 1990; Fig. 1a), limiting the inferential power of any detected relationships, an issue highlighted by Cook (2009) and Kipfmueller et al. (2012) in the context of disagreement between PDO reconstructions.
The use of these particular positive and negative regimes is of specific concern because significant changes are documented in so many components of the climate system between the mid-1970s and the present, as suggested by Cook (2009) in regards to global warming. Since the late 1970s, late summer sea ice extent in the Arctic has decreased by at least 30%, and the rate of ice loss appears to have increased since the early 1990s (Stroeve et al. 2012). The Arctic Oscillation (AO) index shifted from largely negative to very positive in the late 1980s, with recent years displaying very high interannual variability (CPC 2013; Fig. 1b), even as climate models project generally more positive AO values (Gillett and Fyfe 2013). The index of the winter Pacific–North America pattern became positive in the mid-1970s and has largely remained so since that time (Abatzoglou 2011). The Northern Hemisphere storm tracks have shifted northward since the early 1980s (Bender et al. 2012), and they increased in intensity in the early 1970s (Lee et al. 2012). The Atlantic multidecadal oscillation (AMO) index shifted from negative to positive in the mid-1990s (Fig. 1b); variability in Atlantic SSTs also appears to influence North American climate (Alexander et al. 2014) and may mediate the impact of Pacific SSTs on drought (McCabe et al. 2004).
It has been well documented that teleconnections are not necessarily stable over time and that they vary as background conditions change. Gershunov and Barnett (1998) described variability in the strength of ENSO teleconnections that they attributed to shifts in midlatitude Pacific SSTs. Cole and Cook (1998) showed that it was possible for regions to display ENSO teleconnections during some periods but not others and even for the sign of teleconnections to reverse. McCabe et al. (2004) demonstrated that drought response to the PDO over the contiguous United States varied with the state of the AMO. More recently, Wise (2010) reported that the strength of ENSO teleconnections in the western United States may be moderated by conditions in the Atlantic Ocean as well as the Pacific. Moreover, it appears that ENSO conditions can mediate climate anomalies associated with the PDO, and that individual years with similar PDO indices can display variable climate patterns (Wise 2014). The loci of warming in the United States associated with positive phases of the PDO vary with the phase of the AO, as well (Budikova 2005). Lee et al. (2012) found that relationships between the storm track and several modes of climate variability were not stationary, providing a potential mechanism for instability in the relationships between modes of variability and surface temperature and precipitation.
Nor does the paleoclimatic record provide any particular reassurance about the representativeness of PDO teleconnections described using information almost exclusively from second half of the twentieth century. Existing reconstructions of the PDO are not strongly correlated prior to 1900, an observation made by Mantua and Hare (2002), Cook (2009), and Kipfmueller et al. (2012) and shown in Fig. 2. Lack of concurrence between reconstructions could arise from seasonality issues (see St. George et al. 2010) or from uncertainties inherent to statistical reconstruction methodologies, but disagreement between these records could also imply that the regional teleconnections of the PDO may not be stable over time. If reconstructions are based on proxies that respond to local or regional climate, and PDO teleconnections are not robust (i.e., differences in temperature and/or precipitation between the positive and negative phases are not consistent in sign and/or magnitude over time), reconstructions from diverse locations might diverge. This is observed in many PDO reconstructions (Cook 2009; Kipfmueller et al. 2012) from diverse sources including tree rings (Biondi et al. 2001; D’Arrigo et al. 2001; D’Arrigo and Wilson 2006; MacDonald and Case 2005), historical data (Shen et al. 2006), and coral (Felis et al. 2010).
Given the uncertainty as to the cause or causes of the PDO, known temporal variability in many teleconnection patterns, conflicting results from paleoclimate studies, and new knowledge about seasonal variability in the impact of the PDO (Mills and Walsh 2013), evaluating the temporal stability of PDO teleconnections in the observed record is warranted. The occurrence of multiple negative PDO years since a possible regime shift in 1989 from consistently positive PDO conditions to alternating warm and cool conditions (Fig. 1a; Hare and Mantua 2000; Yeh et al. 2011) and the recent development of a long reanalysis (Compo et al. 2011) offer an opportunity to investigate the consistency of PDO teleconnections.
2. Data and methods
Following the methodology outlined by Mills and Walsh (2013), monthly PDO index values were retrieved from the Joint Institute for the Study of the Atmosphere and Ocean. November–March values were averaged to produce an “extended winter” index for the year of the last month (i.e., the 1920 extended winter PDO index is the average of November 1919 to March 1920). To control for the linear influence of ENSO variability on the PDO, a residual PDO index was derived similar, but not identical, to that in Mills and Walsh (2013), using the 5-month running mean Bivariate ENSO Time Series (BEST) index (Smith and Sardeshmukh 2000) to develop a linear regression between the PDO and the BEST index and defining the residual PDO as the difference between the observed PDO index and that predicted by ENSO. While the influence of the tropical Pacific on the PDO may not be strictly linear, this does provide some degree of control. In the present study, the extended winter residual PDO was defined as
where PDO(t) is the extended winter PDO value in year t, and ENSO(t − 1) is the average of the 5-month running mean BEST index values from November through March in year t − 1.
Variability in the previous year’s extended winter BEST index explained about 26% of the variability in the PDO between 1920 and 2011.
Recent PDO index values have not been particularly extreme. To reduce the potential that differences in the subsequent analyses were related to the magnitude of the PDO index, analysis was limited to years when the PDO and rPDO indices fell between ±0.5 and ±2. These ranges are shaded in Fig. 1a, and the years that fall within them are listed in Table 1. Thus this analysis focuses on the climate response in moderately but not extremely positive and negative PDO or rPDO conditions.
PDO impacts on winter [December–February (DJF)] 500-hPa surface heights, sea level pressure (SLP), and surface air temperature were evaluated using the ensemble mean from the Twentieth Century Reanalysis (20CR), version 2 (Compo et al. 2011) available from the National Oceanic and Atmospheric Administration (NOAA) Earth System Research Laboratory. Results were confirmed with the National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) Reanalysis 1 (Kalnay et al. 1996; Kistler et al. 2001); these are discussed only when they differ substantially from the 20CR findings. While there is understandable concern about the quality of the Twentieth Century Reanalysis, particularly as regards high-latitude temperature biases associated with potential errors in prescribed sea ice (Compo et al. 2011), there is also ample evidence that interannual variability in the reanalysis is comparable to other data. Wood and Overland (2010) found that regressions of various climate indices on air temperature and SLP from the first version of 20CR produced results similar to those of other long datasets. While Brönnimann et al. (2012) identified biases and relative weak day-to-day correlations between the Twentieth Century Reanalysis and other datasets, they did find that interannual variability of seasonal average temperature at 1000 hPa was correlated with the Climate Research Unit (CRU) temperature dataset version 3 (CRUTEM3v).
Temperature and precipitation anomalies associated with the PDO were evaluated using the 0.5° × 0.5° monthly temperature and precipitation data from CRU TS 3.2 (Harris et al. 2014) and 0.5° × 0.5° precipitation data from the Global Precipitation Climatology Centre (GPCC) dataset v.6 (Becker et al. 2013; Schneider et al. 2011), which extends only through 2010. Comparing temperature analyses between the Twentieth Century Reanalysis and the CRU gridded observations also provides confirmation of the results (see Figs. 8, 11, and 12). Finally, SST patterns were compared using the NOAA Extended Sea Surface Temperature Reconstruction (ERSST v.3b; Smith et al. 2008).
Annual winter anomalies were defined as a year’s DJF average minus the long-term seasonal mean. Annual precipitation anomalies were standardized by dividing the anomaly by the grid cell long-term mean precipitation. Differences in climate anomalies between corresponding (negative vs negative and positive vs positive) and opposite phases of the PDO were evaluated at every grid cell, using a t test with the assumption of equal variance and without adjusting for temporal autocorrelation.
There has clearly been variability in climate, particularly in Arctic regions (Wood and Overland 2010). Regional or global trends in climate that are unrelated to the PDO could impact composite analysis. To evaluate stability in the pattern of response and control for the influence of long-term trends, correlations between the PDO and climatic variables were performed for 30-yr periods beginning every five years staring in 1920 and ending in 1980 (1920–49, 1925–54, . . . 1980–2009). Performing correlations over 30-yr periods reduces the influence of trends or multidecadal warming unrelated to the PDO that could contaminate the composite analysis presented here. As with the rPDO technique, this methodology may be imperfect, but it can account for regionally varying trends without removing the PDO signal. Similarities between the correlation maps were evaluated by performing pattern correlations between each pair of 30-yr correlations between the PDO and climatic variables. The range of pattern correlations and the contrasts that produced them are shown in Table 2.
Finally, the influences of the northern annular mode (NAM) and Atlantic multidecadal oscillation (AMO) on the expression of the PDO were evaluated. Contrasts between equivalent PDO phases and between negative and positive PDO conditions were evaluated separately during positive and negative NAM and AMO years. Positive and negative NAM years were defined when the Hurrell December through March (DJFM) NAM index exceeded ±0.5 (climatedataguide.ucar.edu/climate-data/hurrell-wintertime-slp-based-northern-annular-mode-nam-index). Positive and negative AMO conditions were defined as years when the DJFM average of the monthly long unsmoothed AMO index was more than 0.1 units from 0 (www.esrl.noaa.gov/psd/data/correlation/amon.us.long.data). These indices are plotted in Fig. 1b, with positive and negative PDO years indicated. Years used in these analyses are listed in Table 1.
3. Results and discussion
a. Sea surface temperature
It was expected that SSTs would largely reflect the canonical pattern, as the EOF analysis would otherwise not indicate positive or negative PDO indices, and that was the case (Figs. 3 and 4). Sea surface temperatures were warm along the western coast of North America and cool in the central Pacific under both positive PDO regimes (Figs. 3a,b), and the opposite was true when the PDO index was negative (Figs. 3c,d). There were, however, critical differences between phases of the same sign, particularly between negative PDOs before and since 1977 (Figs. 3e,f). First, SSTs were between 0.5° and 2°C warmer over the northwest Pacific and across parts of the Bering Sea during recent negative PDO years. The warming was sufficient to degrade the statistically significant difference in SSTs between negative and positive PDO years in this area (Figs. 4a–d). Second, SSTs during recent negative PDO years have been much warmer between 20° and 35°N and 150° to ~200° (Fig. 3f), resembling an extratropical extension of the La Niña pattern. Recent positive PDO years tended to display cooler temperatures in the central Pacific and warmer temperatures in the Bering Sea, suggesting that the recent positive PDO produced more prominent anomalies than positive PDO years prior to the mid-1940s (Fig. 3e); responses to rPDO were qualitatively similar to PDO responses (Figs. 3g,h and 4e–h). Patterns of correlation between the PDO and SSTs were stable, with pattern correlations higher than 0.8 for all comparisons (Table 2).
Recent positive PDO and rPDO years have also displayed cooler western Atlantic SSTs (Figs. 3e,g). This is may be the result of changes in the phase of the AMO. Both positive and negative PDO years tend to display warmer Atlantic SSTs during positive than negative AMO conditions (Figs. 5a,b). There are also differences in SSTs between equivalent PDO phases under different NAM conditions (Figs. 5c,d). However, it is unclear whether the results in Fig. 5 represents a real influence of the NAM on the PDO, as only three years met the criteria for positive PDO/positive NAM conditions (Fig. 1b, Table 1).
b. 500-hPa heights
Positive PDO years are typically associated with a deepening of the Aleutian low. The recent expression has been much stronger (Figs. 6a,b), although some reduction in variance associated with more limited data availability early in the twentieth century (Compo et al. 2011) might contribute to this response. Negative PDO years were characterized by ridging at 500 hPa over the central Pacific during both early and recent negative PDOs, although there were slight differences in the shape, spatial extent, and amplitude of the high (Figs. 6c,d). There were more substantial differences in the pressure patterns over the continent. Prior to 1976, a prominent trough developed over western North America during negative PDO years (Fig. 6d). During recent manifestations of negative PDO, that trough has been substantially weaker and less spatially extensive (Fig. 6c), an effect seen in the correlation analysis as well (not shown). Other recent changes in the PDO include the appearance of a northeasterly trending trough during positive PDO years over the Gulf of Mexico and eastern seaboard of the United States (Figs. 6a,d). Mills and Walsh (2013) saw a similar pressure response in analyses of the NCEP–NCAR reanalysis spanning 1948 to 2007, but earlier analyses (e.g., Mantua et al. 1997) indicate a weak signal in this region. This was also an area where the strength of the correlation between the PDO and 500-hPa heights varied substantially between 30-yr periods (not shown). This may be attributable to Atlantic SSTs, with a stronger response during positive AMO years (Figs. 7a,b).
As with SST, 500-hPa height anomalies were of larger amplitude and spatial extent during recent positive PDO years than in the past, suggesting a weakened pressure differential between the North Pacific and the continent during positive PDO years prior to the mid-1970s. Interestingly, weakening of the western North American trough during negative PDO years and recent enhancement of the positive PDO patterns were more pronounced and spatially extensive when ENSO influence was controlled for by using the rPDO index (Figs. 6e–h). The contrast between positive and negative PDO years was attenuated under positive NAM conditions (Figs. 7c,d). When the positive/negative PDO comparison is limited to positive NAM years, the continental ridge disappears, enhancement of the Aleutian low is quite modest, and the center of the low is displaced southward. This is consistent with the more zonal flow pattern expected when the NAM is positive (Thompson and Wallace 2001), but interpretation is limited by the small number of +PDO/+NAM years and may be complicated by the fact that most −PDO/−NAM years occurred prior to the mid-1970s, whereas −PDO/+NAM years have been more common since that time.
Sea level pressure patterns and differences in the expression of the PDO over time were qualitatively similar to those in 500-hPa heights (not shown); however, there were areas of modest positive (negative) SLP anomalies in the far north underlying negative (positive) anomalies in 500-hPa height over northwestern Canada. Mills and Walsh (2013) also observed this type of response, particularly during December.
c. Surface air temperatures
One source of confidence in statistically detected teleconnections is physical consistency in response across multiple variables. For example, weakening of the Aleutian low should tend to decrease the northward flow of warm air to western Canada and Alaska, leading to cooler temperatures there during negative PDO years. The opposite would be true during positive PDO years, when cyclonic circulation around the Aleutian low draws warm air up through the Pacific toward Alaska and western Canada, and northerly flow east of a ridge over this region encourages cooler than average temperatures over the eastern United States. Temperature anomalies associated with the warm and cool phases of the PDO have been consistent with SLP and 500-hPa anomalies. Positive PDO years were typically associated with warmth over Alaska and northwestern Canada (Figs. 8a,b). During negative PDO years temperatures were, on average, cooler around the North Pacific regardless of when those years occurred, and temperatures were warmer to the southeast (Figs. 8c,d). However, there were important differences in surface air temperature patterns between equivalent phases of the PDO and rPDO at different times (Figs. 8e–h), particularly with regard to warming over Alaska, and these appear consistent with changes in the atmospheric circulation (Figs. 6e–h).
In the southeastern United States, positive PDO years since the mid-1970s were on average up to 2°C cooler than negative PDO years (Figs. 9a,c,e,g), and the differences appear statistically significant, similar to differences in other studies (Budikova 2005; Mills and Walsh 2013). The temperature during positive PDO years between 1920 and 1942 was not distinct from that during negative PDO years, and the difference was often close to zero (Figs. 9b,d,f,h). The recent strengthening of the PDO response over the southeastern United States was also confirmed by correlation analyses (Fig. 10). Neither the Twentieth Century Reanalysis nor the CRU temperature data (not shown) show a widespread statistically significant (p ≤ 0.01) 30-yr correlation to the PDO over the southeast until 1940–69. One potential explanation for this difference is that Atlantic SSTs could influence the response. Indeed, the temperature difference between positive and negative PDO years in this region is more pronounced under negative AMO conditions, when SSTs are cooler, than during positive AMO years (Figs. 11a,b,e,f).
There were also differences over the southwestern United States. Although neither contrast was statistically significant, recent positive PDO years have been warmer than negative PDO years (Figs. 9a,c,e,g), but positive PDO years prior to 1943 were slightly cooler than negative PDO years (Figs. 9b,d,f,h). Correlations tend to show a fairly consistent positive relationship between southwestern temperatures and the PDO, although the strength of the relationship varies over time (Fig. 10). This suggests that unrelated warming may be influencing the composite analysis. The phase of the NAM may also influence the expression of the PDO, with positive PDOs warmer than negative when NAM conditions are negative, and positive PDO conditions cooler when the NAM is positive (Figs. 11c,d,g,h). Neither contrast, however, was statistically significant (p > 0.05).
The truly striking differences, however, were in the north. The most prominent differences between negative and positive PDO years arose from contrasting positive PDO years since the mid-1970s with negative PDO years between 1943 and 1976 (Figs. 9c,g). In all other comparisons, the area over which negative and positive PDOs had a statistically distinguishable temperature difference, particularly over land, was smaller (Fig. 9). In parts of the north, areas with a statistically significant warming response to the PDO detected by comparing positive PDO years between 1977 and 2011 with negative PDO years between 1943 and 1976 showed essentially no response or even nonsignificant cooling associated with positive–negative PDO contrasts over other periods (Figs. 9a,b,e,f). This was seen in 30-yr correlations, as well, suggesting that not all of the weakening in temperature response to PDO in the north was an artifact of Arctic amplification, which could impact the composite analysis (Fig. 10). Similar responses were seen when the comparison was made with rPDO years (cf. Figs. 9 and 12), as well, suggesting that differences in tropical Pacific conditions were not a major contributor to recent attenuation of the northern warming response to positive PDO conditions. Temperatures in the NCEP1 reanalysis showed smaller discrepancies in temperature between recent and midcentury negative PDO years at the highest latitudes (not shown); however, there may be problems with the NCEP1 near-surface temperatures around the turn of the century, related to a change in sea ice data assimilation (http://www.esrl.noaa.gov/psd/data/reanalysis/problems.shtml, accessed 7 February 2014).
The phase of the AMO appeared to have some influence on the strength of the PDO response over eastern Canada (Figs. 11a,b,e,f), although there are differences in the response between the reanalysis and the CRU product (cf. Figs. 11a and 11e). Over western North America, the PDO response was slightly attenuated and shifted farther south during positive AMO years than during negative ones (Figs. 11a,b,e,f). The impact of the NAM on PDO response is more dramatic, with relatively little temperature difference in the north between the two phases of the PDO when the NAM is positive and a strong response when the NAM is negative. Possibly, cooling in the region associated with the positive phase NAM is sufficient to damp any expression of the PDO. This cannot be fully ascertained, however, given the small number of years when both the PDO and the NAM are positive (Fig. 1b; Table 1).
An additional reason for the loss of distinction in temperature between the negative and positive phases of the PDO in the far north may be that early warming in the region was, in fact, not driven by the PDO. Correlation and composite analyses, used in many PDO analyses (e.g., Hartmann and Wendler 2005; Duffy et al. 2005; Mantua and Hare 2002), are useful in detecting differences, but may not be especially effective in determining the cause of those differences. Cassano et al. (2011) investigated warming in northern Alaska in the context of changing circulation pattern frequencies. They found that changes in the synoptic-scale circulation associated with the shift from cool to warm PDO conditions in the mid-1970s should have induced cooling in northern Alaska and thus concluded that warming in the region was unrelated to shifts in the PDO, even though they were contemporaneous. It is also possible that temperatures in the region do have an association with the PDO, but that the relationship was altered by the recent reductions in sea ice. In northern Alaska, sea ice concentration and temperature are strongly correlated (Wendler et al. 2010).
As in Mills and Walsh (2013), this study found that precipitation responses to the PDO were less spatially consistent than those of temperature and pressure, and responses were often relatively weak. As a result, correlation maps with the PDO and rPDO (Figs. 13 and 14) are presented rather than composites. Responses were similar whether precipitation from CRU TS3.2 or GPCC v.6 was used, so only GPCC data are shown, although there were some differences between the two datasets at high latitudes. Positive PDO and rPDO years were typically wetter than average in the along Alaska’s southern coast, and drier over Canada as well as much of the eastern United States, but correlations were not statistically significant over broad areas, and they were not particularly stable over time (Figs. 13 and 14). The strongest responses were observed between 1940 and 1989, with very small areas displaying statistically significant (p ≤ 0.01) correlations after 1980. Over Mexico, precipitation was positively correlated with both the PDO and the rPDO with the exception of 1920–49, when correlations with both indices were mixed (Figs. 13 and 14a). Correlations between precipitation and the PDO over the southwestern United States were generally weak but positive, with the strongest correlations between 1930 and 1969 (Figs. 13b,c). The rPDO was also positively correlated with southwestern U.S. precipitation until 1950–79 (Fig. 14), at which point the correlation became weakly and nonsignificantly negative.
Weakening of the Aleutian low would generally reduce the advection of moisture to the southwestern United States and suppress precipitation in that region. Thus, the relative weakening of PDO-induced precipitation anomalies in that region is consistent with recent attenuation of the 500-hPa height anomalies. Since 1950–79, precipitation has actually been negatively correlated with the rPDO, in contrast to the PDO. The generally weaker and less consistent correlations between precipitation and the rPDO in this region in contrast to the PDO suggest that at least some part of the precipitation signal actually derives from ENSO and not directly from the North Pacific. This is entirely unsurprising given the strong links between precipitation and ENSO in the Southwest (e.g., Cole and Cook 1998) and the potential long-term instability in the relationship between North Pacific SSTs and precipitation in the region (McCabe-Glynn et al. 2013).
e. Comparison to similar studies
Dai (2013) evaluated precipitation anomalies over the contiguous United States for an earlier (1924–45) positive phase of the interdecadal Pacific oscillation index (IPO), which Dai (2013) defines at the second EOF of global SSTs between 60°S and 60°N. The spatial loading of the IPO on SSTs and its time series of variability are similar to the PDO. The responses in annual precipitation during the 1924–45 period were similar to those between 1977 and 1998 (see Dai 2013, his Fig. 5), suggesting reliability in the sign, if not always the magnitude of precipitation response to the positive IPO. There were more differences between the two negative IPO phases, particularly changes in the sign of response over the southeastern United States and the Pacific Northwest (Dai 2013). Northward extension of dry anomalies associated with recent negative PDO years over the Pacific Northwest were also seen in this study (Fig. 13).
In 2004, Deser et al. compared temperature, SST, SLP, and precipitation anomalies across two positive (1900–24 and 1947–76) and two negative phases (1925–46 and 1977–95) of the North Pacific Index (NPI), an index of winter SLP variability in the extratropical Pacific that tracks the strength of the Aleutian low and, like the IPO, has a time series similar to the PDO. Positive NPI phases correspond roughly to negative PDO phases. While their study did not control for individual positive and negative years, they also found consistency in the pattern of winter anomalies: generally cooler conditions to the north and drier conditions, particularly in the southwestern United States, during positive phases of the NPI. As in the present study, however, close examination of the difference figures demonstrates differences in the magnitudes of the responses in different comparisons. There are also subtle differences in the shape of the SLP contrasts that are remarkably similar to those found in this study (not shown).
This paper highlights a significant but unavoidable weakness in earlier analyses of PDO impact on North American climate: that the analyses were typically limited to positive and negative PDO years that characterized one long-term generally positive PDO period and one long-term primarily negative phase. Moreover, many of the positive PDO years were drawn from a time period characterized by a number changes within the climate system that may not have been driven by the PDO but could and likely did impact climate in the Pacific–North American region. These include the reductions in sea ice extent documented by Stroeve et al. (2012), the shift toward more positive AMO conditions, and changes in storm track position (Bender et al. 2012) and intensity (Lee et al. 2012). The current study shares this weakness in that it samples positive and negative PDO years from a relatively small number of distinct time periods.
It is also not clear that the periods from which negative PDO years were drawn are equivalent, either in terms of background climate conditions, as discussed in the introduction, or in their forcing. Alexander (2010) synthesized several decades of research to support the hypothesis that the PDO is not a primary mode of variability, but integrates the impact of multiple potentially interacting climate processes, including midlatitude atmospheric influences on SSTs, tropical coupled atmosphere–ocean variability and midlatitude ocean circulation. Yeh et al. (2011) used observations and modeling studies to show that PDO “transitions” in the mid-1970s and late 1980s derive from different sources and have different atmospheric signatures, even though the SST response is similar in the first EOF. The present study demonstrates differences in 500-hPa height and SLP that are similar to those seen by Yeh et al. (2011) in comparing SLP from 1977–88 with periods before and after. As noted by Alexander (2010), the atmosphere can induce SST anomalies in the North Pacific. Thus, the changes in SLP and 500-hPa geopotential height noted here and by Yeh et al. (2011) could contribute to differences in SST anomaly patterns between similar PDO phases. Changes in atmospheric circulation could also explain why the period from the late 1980s through the present has been characterized by a “mixed” PDO regime, with relatively frequent transitions between warm and cool conditions, in contrast to more consistently warm (1920–42 and 1977–89) and cool (1943–76) periods (Fig. 1a). A different balance among the many possible drivers of the PDO could contribute to changes in the PDO’s temporal autocorrelation structure, as well as to its climatic impacts.
The present study is also somewhat limited in that it evaluates only two periods of positive PDO years and uses negative PDO years from one negative phase and one period (1977–2011) where there is debate about the timing and even existence of a phase shift. It is possible that analysis of additional data would find further differences, identify particular positive or negative periods as unusual, and/or allow us to identify the cause or causes of teleconnection variability. However, at this time it is unclear how a more thorough analysis of spatially continuous observational data could be done, particularly at the most northern latitudes where data quality and coverage are of concern.
One additional weakness of composite analyses is that trends or shifts in climate unrelated to the PDO could influence the magnitude or even sign of the differences, when the relationship is tested over long periods. The correlation analysis (Fig. 10) demonstrates that the basic spatial pattern of temperature response to the PDO is relatively stable, but the strength of that relationship can vary. Variability in the strength or degree of signal is of concern because when statistical relationships between the PDO and modern climate variables are developed for paleoclimatic reconstruction, they assume some degree of stability in the magnitude as well as the sign of response. Other variables, such as precipitation, may also show inconsistency in the sign of response to the PDO at different times, further complicating the evaluation of its teleconnections.
Normally paleoclimate data are considered helpful in determining the impact and robustness of low-frequency climate variability, but in the case of the PDO, disagreement among reconstructions poses a significant problem (Fig. 2; Kipfmueller et al. 2012). In the current study, the potential impact of secular trends and three modes of variability (ENSO, AMO, and NAM) were evaluated statistically, but modeling studies, similar to the ENSO teleconnection analysis by Coats et al. (2013), will be required to evaluate the stability of North American teleconnections to the PDO and identify the mechanistic causes of any inconsistencies. Yu and Zwiers (2007) analyzed climatic response to “in-phase” and “out-of-phase” PDO and ENSO events in a 1000-yr run of the Canadian Centre for Climate Modelling and Analysis (CCCma) first-generation Coupled Global Climate Model (CGCM1) and demonstrated that there can be substantial differences in the response to the PDO, depending on conditions in the tropical Pacific. Furtado et al. (2011), however, identified differences between observations and simulations of Pacific oceanic variability and its relation to the atmosphere that complicate the use of models in understanding variability in the PDO and its teleconnections. Stoner et al. (2009) also found that, although many GCMs simulate SST patterns similar to the PDO, most models have subtle differences. In several models, the leading mode of Pacific SST variability did not resemble that in observations or reanalysis. While confirming that many GCMs did reproduce a SST feature resembling the PDO, Park et al. (2013) found very different correlations between the PDO and various components of the climate system believed to influence the PDO across an ensemble of Coupled Model Intercomparison Project phase 3 (CMIP3) models. Thus, any GCM-based analysis of the PDO will require careful evaluation of model dynamics and, given the low-frequency nature of the variability, either long simulations and/or large ensembles to determine how robust climate variability associated with the PDO is and what influences changes in those teleconnections from one negative or positive phase to the next.
4. Summary and conclusions
This study evaluated the stability of SST, SLP, and geopotential height patterns associated with moderately positive and negative PDO years between 1920 and 2011 and the consistency of temperature and precipitation teleconnections associated with that variability. Although the broad-scale patterns were similar, there were also striking differences in pressure, temperature, and precipitation anomalies between PDO years of the same sign during different portions of the twentieth and early twenty-first centuries. Variability in the magnitude and sometimes the sign of the temperature response north of about 55°N was particularly notable. These distinctions clearly demonstrate some degree of instability in the PDO’s association with climate. The variability cannot be fully explained as an individual function of long-term secular trends, by a linear impact of the tropical Pacific, or by the individual influence of the AMO or NAM, but may be a function of all of these factors, as well as of mechanisms (such as changes in sea ice extent) that this study did not explore. The variability in teleconnections found here is not surprising, as fluctuations in the spatial extent and magnitude of teleconnections to more clearly robust modes of climate variability, such as El Niño–Southern Oscillation, are well documented in observations and climate model simulations (e.g., Cole and Cook 1998; Coats et al. 2013).
Results from this study confirm that there is variability in the climate associations of lower-frequency modes of variability and lead to three critical conclusions. First, given the relatively short historical record, it may be impossible to unambiguously determine the amplitude of climate variability associated with low-frequency variability in the climate system from observations and reanalysis alone. We should, therefore, be particularly cautious about drawing conclusions in relation to low-frequency variability from composite or correlation analysis, even if those inferences seem well supported by internally consistent responses from multiple climate variables. If we cannot be confident in our inferences about the impact of the PDO on climate, it should have limited use in management decisions, as suggested by Wise (2014). Second, the instability of climate responses to the PDO provides a reasonable explanation for disagreement in proxy-based reconstructions of the index and raises significant questions about the wisdom of using such reconstructions in the analysis of past hydrological or ecological variability, a concern identified by Kipfmueller et al. (2012) and Coats et al. (2013). Finally, if the relatively short observational period does not provide an adequate characterization of PDV and/or its links to climate, it may further complicate the assessment of climate model quality over regional and decadal to multidecadal time scales.
PDO reconstructions were downloaded from the NOAA Paleoclimatology Program Archives at http://www.ncdc.noaa.gov/data-access/paleoclimatology-data. SST data and the NCEP–NCAR reanalysis were retrieved from NOAA’s Earth System Science Laboratory at www.esrl.noaa.gov/psd/data/gridded/. Temperature and precipitation from the Climatic Research Unit were made available by the British Atmospheric Data Center (badc.nerc.ac.uk). Precipitation data from the Global Precipitation Climatology Centre were downloaded from their website at gpcc.dwd.de/. The author would like to thank S. St. George, P. Duffy, J. Walsh, J. Littell, and three anonymous reviewers for helpful comments and Tristan Ursell for making the fire.m code available on the Matlab File Exchange.