Industrial-age doubling of snow accumulation in the Alaska Range linked to tropical ocean warming

Future precipitation changes in a warming climate depend regionally upon the response of natural climate modes to anthropogenic forcing. North Pacific hydroclimate is dominated by the Aleutian Low, a semi-permanent wintertime feature characterized by frequent low-pressure conditions that is influenced by tropical Pacific Ocean temperatures through the Pacific-North American (PNA) teleconnection pattern. Instrumental records show a recent increase in coastal Alaskan precipitation and Aleutian Low intensification, but are of insufficient length to accurately assess low frequency trends and forcing mechanisms. Here we present a 1200-year seasonally- to annually-resolved ice core record of snow accumulation from Mt. Hunter in the Alaska Range developed using annual layer counting and four ice-flow thinning models. Under a wide range of glacier flow conditions and layer counting uncertainty, our record shows a doubling of precipitation since ~1840 CE, with recent values exceeding the variability observed over the past millennium. The precipitation increase is nearly synchronous with the warming of western tropical Pacific and Indian Ocean sea surface temperatures. While regional 20th Century warming may account for a portion of the observed precipitation increase on Mt. Hunter, the magnitude and seasonality of the precipitation change indicate a long-term strengthening of the Aleutian Low.


Results
The annually resolved Denali snow accumulation record (Fig. 2) indicates that the post-1950 precipitation increase in the Alaskan weather station records began well before the 20 th century, in circa 1840 CE. The Campbell model produces the smallest accumulation increase since 1840 (Fig. 3), with a doubling of the median accumulation rate between 1600-1840 (0.58 m a −1 ) and 1950-2013 (1.29 m a −1 ), and an average increase of 9.6 mm a −2 over the later period (Figs 2, 3). The first 700 years (800-1500 CE) of the record are characterized by multi-decadal variability and larger uncertainties among our thinning model ensemble that allow for a declining or rising long-term accumulation trend from 800-1500 CE (Fig. 3). Despite the larger uncertainties before 1500 CE, the Denali record shows that accumulation rates from 1950-2013 were the highest of the last 1200 years under all plausible conditions of glacier flow or layer counting uncertainty (Figs 2 and 3). There is nearly complete agreement between the accumulation records in the two Denali ice cores, with a correlation coefficient of 0.92 (p < 0.001). Here, and throughout the text, p-values are adjusted to account for the reduced effective sample size resulting from autocorrelation.
From 1867-2013 CE (133 m depth), the Denali accumulation record can be confidently separated into summer (May-August) and winter (September-April) accumulation due to the high sampling resolution and thick annual layers (Fig. 2 inset). This summer-winter parsing is performed using spring peaks in magnesium (from dust) and late summer-fall peaks in methanesulfonic acid (MSA), which have consistent offsets from other seasonal indicators including ammonium (summer) and δ 18 O (summer peak, winter trough; see Methods; Fig. 1). Figure 2 inset shows that the recent doubling of annual accumulation is dominated by a wintertime increase of 3.6 mm a −2 since 1867, representing an increase of 117% over this interval. The increase in Denali winter accumulation is highly significant (p < 0.001) in both Denali ice cores based on comparison of the winter means from 1870-1900 and 1980-2010, and a non-parametric Thiel-Sen robust linear regression 22 .
Winter Denali accumulation is highly and significantly correlated with winter precipitation at several southern coastal Alaska weather stations, including Valdez, Yakutat, Seward, Juneau, Homer, and Kodiak (Table 1, Fig. 4). The significant correlations (p < 0.05) with Alaskan station data provide additional confidence in our  18 O trough and just below a Mg peak. The distance between each vertical dotted line represents one year's snow accumulation (before thinning correction). The position of these years was selected three times by three independent researchers. We delineate summer (May-August) and winter (September-April) seasons by recording the late summer-fall peak positions of MSA (purple circles) and the spring peak positions of Mg (orange circles).
SCIentIfIC REPORTS | (2017) 7:17869 | DOI:10.1038/s41598-017-18022-5 seasonal parsing and overall fidelity of the Denali accumulation record. Furthermore, the south coastal Alaska station records and the wintertime ice core accumulation record all show the distinct 1976/77 step increase in precipitation associated with the PDO transition from a negative to a positive mode 8,9,23,24 (Fig. 4E). However, while  the weather station records show a step increase in precipitation during the late 1970s, the Denali accumulation record shows that winter precipitation continued to rise on Mt. Hunter after the 1976 shift. We revisit this point in the discussion section below.
Summer accumulation in the Denali record has a more muted but still significant (p < 0.001) increase of 1.2 mm a −2 since 1867, or a total rise of 49% over this interval (Fig. 2). Correlations between the ice core record and Alaskan stations during summer months are restricted to nearby stations in South Central Alaska, including Talkeetna, Homer, Bethel, Iliamna, Kenai, and Seward (Table 1). The summer increase also has a different timing from the wintertime record, characterized by an increase between 1905 and 1925 and relative stability throughout the remainder of the record (Fig. 2 inset). Because of the high temporal resolution (≥8 samples/year) during the last century, the timing of the increases in summer and winter precipitation are robust to seasonal partitioning. Central Alaska has warmed by 2-3° since 1950 8 , and higher precipitation is expected from higher temperatures through the Clausius-Clapeyron (C-C) increase in absolute humidity 3 . While likely responsible for some of the precipitation increase on Mt. Hunter, the observed warming is insufficient to account for most of the doubling of precipitation recorded in the Denali ice core, even with enhanced (14% K −1 ) C-C scaling rates over mountainous regions 25 .
It is very likely that that a strengthening Aleutian Low is responsible for the majority of the accumulation increase observed in the Mt. Hunter ice core. This is supported by the strong relationship between winter Denali accumulation and winter precipitation recorded in weather stations along the Gulf of Alaska, which are sensitive to changes in Aleutian Low strength 26 . Wetter winters in the Mt. Hunter ice core record between 1910-2010 were significantly associated with a deeper Aleutian Low at the surface (p < 0.05; Fig. S4) and 500 mb level (p < 0.1; Fig. 4A) in NCEP/NCAR 20 th century reanalysis data 27 . The 500 mb anomalies include a downstream enhanced ridge over western North America and deeper low over the southeast United States characteristic of the PNA positive teleconnection pattern 28 (Fig. 4A-B). The correlation between the PNA index and winter Denali accumulation from 1950 to 2010 is r = 0.64 (p < 0.0001; Fig. 4E), consistent with a strong relationship between increasing Mt. Hunter snow accumulation and an intensifying Aleutian Low. A 20 th century strengthening Aleutian Low is also supported by a significant negative trend in the North Pacific Index (NPI), a sea-level pressure index over the Aleutian Low region 29 . The negative trend in NPI index during SONDJFMA is significant to p < 0.02, and during DJF is significant to p < 0.01. Previous work indicates that the strength of the Aleutian Low tends to co-vary with its mean position, with a stronger (weaker) Aleutian Low shifted towards the east (west) 30,31 . However, our observations show that the highest accumulation winters on Mt. Hunter are associated with a sea level pressure anomaly centered over the Aleutian Islands (Fig. S4). Thus, we observe no strong evidence for covariance between Aleutian Low position and Mt. Hunter snowfall.

Discussion
Under all four ice-thinning models, the Mt. Hunter snow accumulation rates in the 20 th century are unprecedented in the last millennium (Figs 2, 3). The Campbell finite element ice flow model 14 produces the highest degree of thinning since 1650 CE among any of the model simulations we used, and therefore is the most conservative model in that it produces the smallest precipitation increase on Mt. Hunter since 1840 CE. This is our preferred ice flow model because it is constrained by the detailed basin and glacier geometry, spatial accumulation patterns, and surface ice velocity of the Mt. Hunter ice core drill site. Prior to 1650, the Hooke model 12 consistently produces higher accumulation values than the other thinning models (Fig. 3). This may be because parameters in the Hooke model are statistically optimized to match the timescale, as opposed to the other three models where thinning corrections are based on explicit calculations of vertical strain rate (see Supplemental Information). Even in the Hooke model, the accumulation rates before 1840 never approach 20 th century values (Fig. 3). Numerous accumulation reconstructions from alpine ice cores have been published during recent decades. Mt. Hunter is very similar to many of these sites with low velocities near the drill site (approximately 2 m a −1 ), surface-conformable stratigraphy, and very low surface slope. The average annual temperature on Mt. Hunter (−17 °C) is cooler than at most other alpine ice core sites but is similar to Ushkovsky volcano 32 (−15.7 °C) and Dasuopu glacier 33 (−16 °C). Many other alpine ice core accumulation records have been drilled at sites with saddle-like geometry including Aurora Peak 16 , Mt. Waddington 21 , Mt. Everest 12 , Belukha Saddle 34 and Mt. Logan 15 . Thus, we would expect that the modeling approaches published in these studies would be applicable to the conditions on Mt. Hunter. Nevertheless, we use four different modeling approaches, one of which is designed for the specific Mt. Hunter site geometry, and all of which have very conservative and rigorously quantified ranges of uncertainty ( Fig. 3 and Supplemental Information). To our knowledge, no other alpine ice core accumulation record has been developed with such a thorough characterization of the thinning regime or uncertainties. All of the thinning models produce a robust increase in accumulation since the mid-19 th century above late-Holocene background values.
While all of our modeling simulations agree that during the early half of our record (810-1400 CE) precipitation remained well below present values, we cannot confidently conclude from our results whether precipitation was increasing or decreasing within this time period. One simulation (the Hooke model 12 ) shows a slight decrease in accumulation rate between 810 and 1400 CE while the remaining three modeling approaches suggest a slight accumulation increase during this period with superimposed decadal to centennial scale variability. Relatively low snow accumulation on Mt. Hunter between 1000 and 1300 CE is consistent with low sodium concentrations in the Mt. Logan and Mt. Hunter records during this interval, indicative of a relatively weak Aleutian Low (Fig. S5) 35,36 .
We suggest that the recent doubling of Alaska Range snowfall recorded in the Denali ice core reflects an increase in absolute humidity with warming and an intensification of the Aleutian Low to its strongest levels (most positive PNA) of the past millennium. The precipitation increase we observe on Mt. Hunter is apparent in other Alaskan records. Between the mid-20 th century and present day, several weather stations along coastal Alaska 6 (e.g. Kodiak, Palmer, Juneau) have experienced similar (but smaller) increases in wintertime precipitation compared with Mt. Hunter (Fig. S6). Consistent with our results, a century-long ice core collected from Aurora Peak in the eastern Alaska Range 16 (240 km from Mt. Hunter) also records a doubling of snow accumulation over the 20 th century, with greater rates of precipitation increase than observed on Mt. Hunter (Figs S5, S6). The Mt. Logan ice core (Yukon, Canada) time series of accumulation, sodium, and water isotope ratios show trends over the past 200 years indicative of a strengthening Aleutian Low 15,35,37 (Fig. S5). Specifically, Moore et al. 15 found a positive and accelerating trend in Mt. Logan annual accumulation after the middle of the nineteenth century, and interpret it as indicative of a progressively more positive PNA. This was supported by Rupper et al. 38 who confirm that high accumulation years on Mt. Logan are associated with a stronger wintertime Aleutian Low. Furthermore, the composite Mt. Logan and Mt. Hunter sea-salt Na + calibrated Aleutian Low proxy similarly supports a progressive strengthening of the Aleutian Low over recent centuries to its strongest levels of the past 1500 years 36 (Fig. S5).
The doubling of accumulation rates at Mt. Hunter, Aurora Peak and other sites near the Gulf of Alaska stand seemingly in contrast to the more modest 2-3% increases in precipitation per degree of warming predicted by climate models (e.g. Held and Soden 3 ). The rapid precipitation increase in Alaska represents a regional signal amplified primarily by enhanced advection of moisture and storms into the Gulf of Alaska associated with a strengthening Aleutian Low. With a deeper Aleutian Low, the largest precipitation increases would be expected to occur where southerly flow on the eastern side of the low encounters the Alaskan coastline and mountain ranges. This is exactly the pattern we observe, with increasing precipitation at coastal Alaskan weather stations including Kodiak, Juneau and Valdez, and the strongest precipitation increases at orographic barriers such as Mt. Hunter and Aurora Peak ( Fig. 4d; Fig. S6). Previous work has shown that this configuration leads to drier conditions within the rain shadow of these orographic barriers 39 , which is apparent as negative correlations between winter Mt. Hunter accumulation and precipitation at Tanana, McKinley Park, and Fairbanks located north of the Alaska Range (Table 1). Additionally, Tanana winter precipitation has experienced a significant decrease since 1950 6 . We do not, therefore, infer that the Mt. Hunter precipitation increase is representative of the expected broader mid-high latitude precipitation increase predicted by global scale climate projections 40 . Rather, the Mt. Hunter record captures a specific regional phenomenon (a deepening Aleutian Low) that has occurred in conjunction with warming, resulting in a dramatic increase in regional precipitation. Finally, ice core 17 and other paleoproxy records 41 from outside the North Pacific region do not document long-term precipitation increases, consistent with our finding that the North Pacific hydroclimate intensification is a regional feature associated with a strengthening Aleutian Low.
Dynamical studies suggest that the most plausible way to intensify the Aleutian Low is through warmer tropical SSTs, particularly in the western tropical Pacific and Indian Oceans 42 . Enhanced convective precipitation and latent heat release associated with higher tropical Pacific and Indian Ocean SSTs induce a Rossby wave extratropical response manifesting as the PNA positive pattern 42,43 (Fig. 4b). Consistent with this dynamical theory, the annual Mt. Hunter accumulation record from 1910-2010 is significantly correlated (r = 0.59, p < 0.01) with reanalysis SSTs in the western tropical Pacific and Indian Oceans (Fig. 5), supporting a strong link between Alaskan precipitation and tropical SSTs. This high correlation is due in part to the highly trended nature of both records, with warming temperatures over the tropical Pacific and Indian Oceans and rising accumulation in the Alaska Range. However, a significant correlation remains after detrending both time series (r = 0.39, p > 0.01), demonstrating the robust nature of the link between the tropical Pacific and Indian Ocean and North Pacific climate identified by previous studies 42, 43 .
In Fig. 6, we compare the Denali accumulation record to multiproxy records of SSTs in the Western Pacific and Indian Oceans 44,45 . Sustained and significant changes since ~1840 CE are evident in each dataset (Denali = 1840 CE, Western Pacific = 1834 CE, Indian = 1827 CE), which we confirm using the change detection methods described in Abram et al. 44 (see Methods). This is consistent with the doubling of Denali snow accumulation since ~1840 being forced, at least in part, by warmer western Pacific and Indian Ocean SSTs through the atmospheric bridge 43 . The secular increase in Denali and Mt. Logan ice core sea salt Na + concentrations over recent centuries has also been interpreted as an extratropical response to warming Pacific SSTs 35,36 , as has a dramatic increase in the Mt. Logan δ 18 O record at 1840 37 (Fig. S5). Similarly, a proxy record of Hawaii precipitation includes a secular change towards more PNA positive conditions, reflected as a long-term precipitation decrease (Fig. 6) associated with anomalously high pressure in the subtropical North Pacific 46 (Fig. 4b). Thus, through the atmospheric  Inspection of Fig. 5b reveals that both the western tropical Pacific-Indian Ocean SSTs and Mt. Hunter wintertime accumulation have continued to progressively increase since the 1976/77 PDO transition. This post-1977 increase contrasts with steady wintertime precipitation at south coastal Alaska weather stations following the precipitation jump in 1976/77 (Fig. 4E). This may indicate a heightened sensitivity to tropical SST teleconnections at higher elevation in Alaska. Other studies from the nearby St. Elias range have inferred a similar sensitivity of high elevation hydroclimate to conditions in the tropical Pacific Ocean 35,37 . The Rossby waves responsible for these tropical teleconnections are best defined in the free troposphere 43 , which may explain the preferential expression of tropical connections at high altitudes in Alaska. Additionally, mountain summit areas are more representative of free-air conditions compared to stations located in valleys or along the coast where local effects may mask the tropical climate signal 47 .
We examined output from seven CMIP5 GCM simulations (Table S1) to assess whether Last Millennium Experiment reconstructions capture the Aleutian Low intensification we infer from the Mt. Hunter record. All of the GCM simulations show warming in the western Pacific (15°S-15°N, 110°-160°E) with an ensemble average warming of 0.57 °C during the last two centuries comparing well with the 0.5° warming estimate from paleoclimate records 45 . In contrast, only 3 of the 7 models show significant (p < 0.05) deepening of the Aleutian Low (defined by the NPI index) (Fig. 6). None of the models match the observed increase in precipitation over the central Alaska Range or southern Alaska (58°-62°N, 135°-155°W), nor does the ensemble show a consistent precipitation decrease in Hawaii (Fig. S7). Those models that do suggest a precipitation increase in southern Alaska (MPI-ESM 48 and IPSL-CM5A-LR 49 ) produce a 5-10% rise in precipitation during the last two centuries, considerably less than shown in the Denali accumulation record. Furthermore, these two models are both among the simulations that generate an Aleutian Low deepening since 1840, and have the greatest increase in Western Pacific SSTs, although the SST increase in both models is significantly larger than in proxy reconstructions 45 . We suggest that current GCMs underestimate the sensitivity of North Pacific atmospheric teleconnections to tropical SSTs, and thus further model development focused on improving the extra-tropical responses to tropical SST warming may improve North Pacific hydroclimate projections in a warming world.
Although the future response of central-eastern tropical Pacific SSTs to increased radiative forcing remains unclear 50 , GCM simulations show enhanced ocean warming in the western tropical Pacific and Indian Ocean through the 21 st century in response to higher GHG concentrations 51,52 . The Mt. Hunter accumulation record corroborates and significantly extends instrumental (Fig. 4) and paleoclimate 15,16 records showing precipitation increases in southern Alaska during the 20 th century. While hemispheric-scale reconstructions have yet to reveal coeval intensification of 20 th century hydroclimate anomalies outside of the North Pacific 41 , our data imply that regions such as Mt. Hunter, which are sensitive to tropical teleconnections, may continue to experience hydroclimate variability well outside the natural range of the past millennium.

Methods
Ice Core Collection and Processing. The Mt. Hunter ice cores were drilled in 2013 at 3,900 meters elevation from the saddle between the north and middle peaks of Mt. Hunter. Two 8 cm diameter cores were drilled to bedrock (208 m deep) using a Badger-Eclipse ice drill. After drilling, ice cores were shipped to the National Ice Core Laboratory in Denver, Colorado where the cores were weighed and measured for density calculations at ~1 m resolution, and cut into 3 × 3 cm sticks for melting at Dartmouth. Laboratory Procedures. The ice cores were continuously sampled on the Dartmouth Ice Core Melter system, which features an ultra-clean silicon carbide melt head that separates potentially contaminated meltwater from the outside of the core and pristine meltwater from the core interior 53 . The water from the interior fraction was pumped through an Abakus (Klotz) laser particle counter and particle size analyzer, a liquid conductivity meter (LCM) and finally into pre-cleaned vials for chemical analyses 54  Ice core melter samples were collected at a resolution averaging 5 cm, depending on melt rate, except that isotope samples were collected at ~1 cm resolution from 1750 to 1250 CE. LCM and particle concentration and size distribution were measured at an effective resolution of 1 millimeter. Major ion concentrations were measured on a Dionex 5000 capillary ion chromatograph (IC). Each IC analysis run of 50 samples included two sets of anion and cation standards for each analyte as well as four clean water blanks directly from the MilliQ filtration system, two blanks of MilliQ water from the bottle used to flush the lines in the melter system, and three blanks of MilliQ water that has been run through the melter head and the entire system. Water isotope ratios were measured by a Picarro L2130i Cavity Ring Down Spectrometer linked to a high throughput vaporizer at the University of Maine. Results are reported relative to SMOW (Standard Mean Ocean Water). Long-term precision is 0.1‰ (1σ) for δ 18 O and 0.4‰ (1σ) for δD, based on multiple analyses of internal and international standards.
Ice Core Dating. To develop the Denali ice core timescale, three experienced researchers (D.W., E.O., D.F.) independently counted annual layers in the Mt. Hunter ice core chemical data. The timescale to 1777 CE was determined by counting annual oscillations in δ 18 O (summer peak), melt layers (summer peak), magnesium (spring peak), dust (spring peak), liquid conductivity (summer peak), ammonium (summer peak) and SCIentIfIC REPORTS | (2017) 7:17869 | DOI:10.1038/s41598-017-18022-5 methanesulfonic acid (MSA; late summer-fall peak), consistent with previous North Pacific ice cores 16,21,35,[55][56][57] . Using the relative peak positions of MSA in the late summer-fall and magnesium in the spring we distinguish cold-season (September-April) and warm-season (May-August) snow accumulation on Mt. Hunter back to 1867 CE. The timescale from 1777 to 1500 CE is based on annual oscillations of δ 18 O, δD, deuterium excess, dust and liquid conductivity measurements that were made at higher resolution than the other analytes, while conductivity and dust concentration measurements were exclusively used to date the ice core from 1500-800 CE.
The Mt. Hunter chronology is validated from 1750-2013 CE by comparing the timing of peaks in sulfate, chloride and conductivity to the known dates of explosive volcanic eruptions 53,58,59 (Fig. S2). The volcanic events are used only as validation of our layer counting efforts and the timescale was not forced to match any particular event. There is no offset between our timescale and known volcanic eruptions as indicated by peaks in sulfate, chloride and conductivity during the 19 th and 20 th centuries, indicating a precision within +/−1 year throughout the last 200 years. The volcanoes in the 18 th century used for chronology validation (Laki, 1784 and Pavlof, 1763) were offset by one year from our chronology. Additionally, Cs-137 concentrations in the Mt. Hunter core strongly peak in 1963 during the largest atmospheric nuclear weapons testing campaign, which closely matches other published data 16,56 . Accumulation Uncertainty. Errors in the reported accumulation values (Y direction in Fig. 2) are due to three factors. First, there is uncertainty due to spatial and stochastic processes including wind redistribution, which we quantify using the difference in layer thicknesses between the primary and secondary ice cores. The layer thickness records from the two cores have a correlation coefficient of 0.92 and a slope of 0.96. During the last 100 years, the average percent difference in annual accumulation between the two cores is 10.2%, which we assume is consistent throughout the core and which we refer to as "stochastic error".
There is a range of depths for each January 1 position because the three layer counters picked slightly different spots due to abnormal peak shapes, ambiguity, dispersion and human error. For each annual layer, the so-called "Peak Position Error" is quantified by dividing the range in January 1 position by the annual layer thickness. In cases where only two of the three counters picked a year, we assign an uncertainty of +/−50% of the annual layer thickness to that year. In cases where the core was damaged and equally spaced years were assigned, the uncertainty range is defined as the difference between the thinnest and thickest layers within 10 years of the gap. Peak position error is generally equal to about 10 cm (w. eq.) in the final thinning-corrected accumulation record, although it is occasionally larger in sections of low core quality when uncertainty assignments of +/−50% are more common.
As described in detail in the Supporting Information, we use four ice flow models to quantify the level of uncertainty in accumulation rate arising from the choice of flow model for layer thinning correction. For each year we define the "modeling error" as the entire range between the minimum and maximum model runs.
Stochastic error and peak position error are assumed to be symmetrically distributed around a single mean accumulation value and are aggregated using Eq. 1, where U t is the total uncertainty in year t, S t is stochastic uncertainty and P t is peak position uncertainty. We then define our upper and lower uncertainty bounds by adding U t to the maximum uncertainty bound from our flow modeling simulations, and subtracting U t from our minimum flow modeling uncertainty bound. This analysis shows that it is virtually certain that accumu lation on Mt. Hunter has never approached its present day values at any time in the last millennium.

Analysis of Instrumental and Reanalysis Data.
Alaska weather station data were obtained from the Global Historical Climatology Network 6 and accessed through the Climate Change Institute Climate Reanalyzer. Months with missing precipitation data were filled with the average value for that month over the period of record. Reanalysis datasets were accessed through the Climate Change Institute Climate Reanalyzer website as well as the NCEP/NCAR monthly composites website. We use NCEP/NCAR 20 th century reanalysis data 27 because of the long dataset duration. We restrict our analysis to the last 100 years (1910-2010) since both the resolution of our seasonal accumulation record and the reliability of the reanalysis data are higher after 1910.
Change Point Detection. Following Abram et al. 44 , we define the point of sustained and significant increase in our time series as the point at which trends significant at p < 0.1 persist through present day. We used SiZer analysis with code from Abram et al. 44  GCM output of sea level pressure and precipitation rate over the Alaska region over the duration of our study. GCM data were accessed through the Earth System Grid Federation data node https://pcmdi.llnl.gov/esgf-idp. There were seven GCMs available with data spanning the period from 1000-2005 CE and each of these is described in Table S1.
SCIentIfIC REPORTS | (2017) 7:17869 | DOI:10.1038/s41598-017-18022-5 We calculated the average monthly sea level pressure values over the area from 160°E to 140°W and 30°N to 65°N in order to reconstruct the North Pacific Index (NPI) as defined in Trenberth and Hurrell 28 . We used GCM output from September-April to match our winter accumulation record. Sea surface temperature is not available across all seven GCM ensembles so we use surface air temperature from 15°S-15°N, 110°-160°E as an indicator of surface conditions over the Western Pacific warm pool. Results are standardized relative to the 1980-2000 mean. We extracted monthly precipitation rate output from the GCMs over the Western Pacific (15°S-15°N, 110°-160°E), Hawaii (15°-25°N, 155°-170°W), Southern Alaska (58°-62°N, 135°-155°W) and Mt. Hunter (63°N, 151°W), and combined the data into annual and seasonal averages (Fig. S7).