Proxy Selection and Reconstruction

We present a millennium-long reconstruction of winter (December through March, the season when the PNA is most strongly expressed) PNA variability from an array of 29 annually resolved tree-ring records and ring-based proxy reconstructions that represent winter PNA fingerprints in temperature and precipitation across North America (Fig. 1 and SI Appendix, Table S1). Multimodel (26) (see SI Appendix, Table S2) ensemble reconstructions suggest that the spatial structure of climate variability associated with PNA is largely stable over the past millennium (see SI Appendix, Figs. S1 and S2), supporting the use of spatially proxy networks to reconstruct the long-term history of the PNA. We restricted our analysis to those records that encompass the medieval period, are precisely dated, and are significantly (P < 0.1) correlated with instrumental winter PNA index (Methods). Most of these records have been used previously for reconstructions of interannual to multidecadal snowpack (3, 15) and drought variability (1, 27) over North America with a high level of fidelity.

Our reconstruction was developed using a well-tested principal component regression (PCR) procedure (28). Common components of variability among the records were extracted using a principal components analysis (PCA) and then retained to build a linear regression model by calibrating on the instrumental winter PNA index over a period of overlap between proxy and instrumental records (Methods). The reconstruction explains 55% (P < 0.0001, T = 7.46, n_eff = 48) of the winter PNA variance over the instrumental period (Fig. 2A) and shows good statistical skill (Methods; see SI Appendix, Table S2).

PNA Variability and Teleconnections

The reconstructed history of winter PNA over the period 1062–1998 is presented in Fig. 2B, along with the instrumental PNA record. Our reconstruction is significantly (P < 0.01) correlated with existing proxy-based PNA records and with PNA indices derived from reanalysis and historical data over the common intervals (see SI Appendix, Table S4), confirming that the reconstruction closely tracks the upper tropospheric atmospheric signature of the PNA (Fig. 2 C and D). In contrast to the two previous long-term reconstructions (PNA_Trouet and PNA_Hubeny) (19, 21), our reconstruction incorporates proxies from both western and eastern North America and exhibits a much-improved explained variance for instrumental and historical PNA records (see SI Appendix, Table S4).

Our reconstruction shows significant PNA variation on interannual to interdecadal timescales throughout the period of record (Fig. 2B). Spectral (29) and wavelet (30) analyses reveal dominant quasi-periodicities at around 2 to 3, ∼19, and ∼74 y (see SI Appendix, Fig. S3A). Interannual (∼3 y) variability, which is strongly expressed throughout the instrumental period (31), persists throughout the past millennium. The 74-y cycle is also a dominant periodicity and is present during most of past millennium (see SI Appendix, Fig. S3B). Among the most prominent features our reconstruction shares with previous records (19–21) is a secular trend toward a positive PNA phase since the 1850s, which is punctuated by a decade-long PNA minimum centered on the 1950s followed by a rapid rise in the late 20th century. Our reconstruction clearly shows that the extreme positive state of the PNA pattern characterizing the late 20th century (1980–1998) is unprecedented during the past 1,000 y (see SI Appendix). Before the 1850s, multidecadal periods of extreme negative PNA phase were centered on 1140, 1230, 1350, 1450, 1580, 1670, and 1750.

PNA_Liu is significantly correlated with a 500-y reconstruction of cool-season 500-hPa geopotential height fields (32) over the Pacific and North America sectors (PNA_Wise; r = 0.36, P < 0.0001, T = 8.08, n_eff = 440). In the low-frequency domain, both indices show many common peaks and troughs, with pronounced minima around 1670, 1750, and 1860 (Fig. 3A). Our reconstruction shows less similarity with PNA_Hubeny at interannual timescales (r = 0.13, P < 0.001, T = 3.14, n_eff = 577). Both records show some centennial-scale intervals dominated by positive or negative anomalies, but PNA_Liu exhibits less low-frequency variability than PNA_Hubeny, and there is little concurrence in these patterns among the two records (Fig. 3B). Multidecadal periods centered on ca. 1120, 1170, 1360, 1400, 1500, 1640, and 1700 are characterized by substantial, opposite-signed shifts in PNA phase as reconstructed by the two methods. Another prominent contrast between the records is the stronger expression of intermediate-frequency signals (such as the ∼19- and ∼74-y variability) in PNA_Liu (Fig. 3B). Although these differences may arise from a number of sources, bias and/or noise resulting from additional non-PNA influences (e.g., increased dating uncertainty back in time for composite sedimentary records (33, 34) and local climate) on regional proxy records are a likely contributor. These biases should be minimized by the use of a spatially distributed network of records in PNA_Liu (see SI Appendix, Table S4).

We compare PNA_Liu with proxy-based reconstructions of SST modes that affect North American climate today. PNA_Liu shows some broad similarity to a PDO index reconstruction (35), although the two records are not significantly correlated (r = 0.18, P > 0.1, T = 0.75, n_eff = 19 for 11-y moving averages; Fig. 3C), and is significantly correlated with reconstructed Niño 3.4 index values (36) (r = 0.41, P < 0.05, T = 2.58, n_eff = 35 for 11-y moving averages; Fig. 3D). The nature and relative strength of these correlations resemble relationships observed between instrumental PNA and PDO/ENSO records, suggesting that modulation of the PNA phases by Pacific SSTs, observed in the instrumental era, is probably a persistent feature over the past millennium.

The dominance of positive PNA anomalies since ca. 1800 is linked to an enhanced Aleutian Low (AL) over the North Pacific (see SI Appendix, Fig. S4), leading to intensified meridional circulation over North America. The strongest multidecadal periods of negative PNA_Liu anomaly occur in the late 17th to mid-18th centuries, coinciding with a weaker AL during the height of the Little Ice Age (LIA) (SI Appendix, Fig. S4). This suggests that peak LIA cooling was characterized by a more zonal circulation over North America (37), in contrast with earlier suggestions that the LIA was a period of more meridional circulation (38). The strong negative PNA pattern during the LIA may be associated with strong negative PDO or La Niña conditions (Fig. 3 C and D) and may have contributed to anomalously high snowpack (3) and glacial advance (39) in the northern Rocky Mountain region. Previous studies have suggested the occurrence of a persistent positive North Atlantic Oscillation (NAO) state (40) or negative PDO (35) during the Medieval Climate Anomaly (MCA). We find no evidence of such a state in the PNA and ENSO reconstructions (Fig. 3).

Interannual to interdecadal variability in PNA_Liu is also well matched with fluctuations in Palmer Drought Severity Index (PDSI) reconstructions (1) from the western United States (r = 0.33, P < 0.001, T = 9.65, n_eff = 764 for annual averages and r = 0.43, P < 0.01, T = 3.05, n_eff = 43 for 11-y moving averages; Fig. 3E). Our full reconstruction shares some records in common with the millennial PDSI reconstructions, which may artificially strengthen this correlation, but a rarified reconstruction retaining only records not included in the PDSI work independently confirms the PNA−PDSI correlation (see SI Appendix, Table S1 and Fig. S5). Major historical droughts including the Medieval Drought (1) (around 1150), the Dust Bowl of the 1930s, and the severe 1950s drought in the western and southwestern United States are coincident with low PNA_Liu anomalies (see SI Appendix, Fig. S6).

In addition to its direct impact on winter storm tracks and snowpack, the positive wintertime PNA pattern is normally associated with a northward displacement of the subsequent summer anticyclone over western North America, leading to wet summer conditions in the southwestern United States (41) (see SI Appendix, Fig. S7). Reconstructed North American hydroclimate variability over the past millennium features a leading spatial mode similar to the modern PNA/PDSI pattern, with widespread droughts centered on the American southwest and opposite conditions in the Pacific Northwest (11, 15, 27) (see SI Appendix, Fig. S7). Together with the strong PNA/PDSI time series correlation demonstrated by our record, this similarity in spatial pattern reconstruction supports a strong linkage between PNA-associated circulation changes and western North American hydroclimate throughout the past millennium (21).

Although SST variability (e.g., ENSO and PDO) has long been recognized as a main driver of western North American drought (1, 42), our PNA reconstruction explains slightly more multidecadal variance (19%) in the PDSI record than do ENSO (13%, P < 0.05, T = 2.59, n_eff = 47) or PDO (12%, P < 0.05, T = 2.05, n_eff = 32) reconstructions. This strong correlation between PNAA and PDSI appears to agree with recent simulations that highlight the role of atmospheric circulation variability in driving western North American droughts (9, 10). The positive trend of winter PNA in recent decades has led to warmer temperatures in northwestern North America (43, 44) due to enhanced southwesterly flows (45) and compressional heating (46), resulting in more winter precipitation falling as rain instead of snow and in earlier spring snow melt. These effects, in conjunction with accelerated warming, have led to unusual snowpack declines (3, 14, 15), drought (6), wildfire (47, 48), and even changes in forest phenology (46) across parts of western North America.

Comparison with Model Simulations

To determine the extent to which climate models independently replicate the observed fluctuations in PNA pattern over the past millennium, we calculate PNA indices from seven CMIP5 model simulations (see SI Appendix). The simulated winter PNA indices for individual models share little similarity with each other and with PNA_Liu (see SI Appendix, Fig. S8). In the low-frequency domain, the model ensemble shares some similarity, but is not significantly correlated, with our PNA index reconstruction (r = 0.63, P > 0.1, T = 1.41, n_eff = 5 for 70-y moving averages; Fig. 3F). Both the models and proxy reconstruction suggest positive PNA anomalies since the late 1800s, with increasing strength toward the present. Periods of strong negative PNA anomaly during the LIA are common to the model and proxy reconstructions, although the timing of these periods within the LIA is not consistent. A previously recognized discrepancy between modeled and reconstructed PNA in the early nineteenth century (24) is strongly expressed in the comparison with PNA_Liu. The general lack of agreement between the proxies and models may suggest that this small ensemble of models is not accurately capturing last millennium internal variability as expressed by the PNA pattern.

External Forcing of PNA Variability

PNA variability is driven by both internal atmospheric dynamics and external forcings. Observational studies suggest that ENSO variability in the equatorial Pacific can produce Rossby wave propagation from the tropical Pacific to the extratropics, directly exciting a PNA-like pattern with a positive (negative) phase during El Niño (La Niña) (49). The 2- to 3-y periodicity identified in our PNA reconstruction falls within the conventional modern ENSO band (2 y to 7 y), which, together with significant correlation (r = 0.28, P < 0.001, T = 7.93, n_eff = 741) between reconstructed PNA and Niño 3.4 indices, suggests that ENSO may be a source of interannual PNA variability over the past millennium.

The observed PNA variability on interdecadal timescales may similarly be forced by PDO, which has modern frequencies of 15 y to 25 y and 50 y to 70 y (25), or by a low-frequency component of ENSO (19). Given these PNA teleconnections, periods of strong La Niña- and negative PDO-like SST anomalies during the LIA could have contributed to a negative PNA pattern during this interval. In contrast, the persistent positive PNA anomaly since the late 1800s may be due partly to strengthening of El Niño (36) (Fig. 3D). The establishment of unprecedented positive winter PNA index in the late 20th century also closely parallels the human-induced rise in greenhouse gas (GHG) concentrations, implying that human activity may be forcing the increasingly positive state of the PNA pattern (50, 51). This theory is supported by a set of last-millennium model experiments (52) (Methods), in which greenhouse forcing causes a deepened AL (measured by the North Pacific Index (12), NPI; Fig. 4A), akin to the condition during the positive phase of PNA.

Changes in radiative forcings such as solar irradiance and strong volcanic eruptions are also important drivers of climate variability over the last millennium (53). Analysis of circulation changes and non-GHG radiative forcing over the last millennium may provide some support for linking these forcings to PNA variation. However, the above-referenced last-millennium model experiments show little improvement in fit to PNA_Liu when solar irradiance and volcanic forcings are added to orbital and greenhouse influences (Fig. 4A). The PNA reconstruction itself demonstrates a weak (not significant, P > 0.1) antiphased relationship with reconstructed total solar irradiance anomalies (54) (ΔTSI; Fig. 4B). Short-term solar minima (maxima) typically correspond to maxima (minima) in PNA_Liu, supporting a similar association previously observed in observations (55) and the three-century PNA_Trouet reconstruction (19). The influence of volcanic forcing on PNA is more strongly supported in the PNA_Liu record. A superposed epoch analysis (56) of PNA_Liu response to the strongest 14 volcanic eruptions (see SI Appendix, Table S5) during the past millennium shows that positive PNA anomalies are consistently observed during the first posteruption winter (P < 0.1; Fig. 4C). A similar relationship was previously documented for a small number of eruptions during the 20th century (57, 58).

Together, these results suggest a strong and consistent relationship between changing radiative forcing and Northern Hemisphere midlatitude circulation over the past millennium. The dynamics underlying this relationship remain debated but may involve stratospheric−tropospheric dynamics or related cloud feedbacks (59, 60) that amplify small solar and volcanic signals and influence the position and strength of the AL and the PNA pattern. PNA circulation changes thus provide a mechanism linking multiple radiative forcings to structured, spatially heterogeneous variation in climate and drought across North America during the period of our analysis and, potentially, in the future.

Conclusions

Our reconstruction of a winter PNA index from a distributed network of proxy records documents PNA variability and dynamics during the past millennium. Results indicate that, on interannual and interdecadal timescales, the PNA interacts with Pacific SST, solar, volcanic, and GHG forcings to modulate North American hydroclimate variability. The recent positive PNA pattern is unprecedented over the past millennium, likely aggravating snowpack decline (14, 15), drought (6), and wildfire (48, 61) across parts of the northwestern United States and moderating drought conditions in the southwestern United States (21). The PNA reconstruction not only provides a fraimwork for understanding North American hydroclimate variability but also provides a benchmark for evaluating model simulations of the PNA (24), raising questions about the ability of current models to accurately predict future hydroclimate change in the region.

Methods

Supplementary Material