Surprising PNAS paper: CO2 emissions not the cause of U.S. West Coast warming

Materials and Methods
The primary SST data for this study consist of monthly 2°-gridded and fully
interpolated fields from the ERSST (23). Primary SLP data come from the
monthly NCAR Northern Hemisphere 5° analysis (24). The NCAR SLP dataset
is a quality-controlled compilation of daily surface map analyses from various
sources. Comparisons with other SST and SLP datasets are illustrated in
Fig. S3.
To objectively isolate large-scale modes of NE Pacific variability, EOF analysis
was first performed separately on normalized monthly matrices of SLP and SST.
The primary modes in each field were then compared, first by correlation, with
statistical significance estimated fromaMonte Carlo simulation using synthetic
random SLP1 series with observed levels of autocorrelation and low-frequency
variance. A stochastic time series model was created to estimate the monthly
SST1 response to a combination of SLP1 forcing and monthly lag − 1 SST
persistence, an approach originally developed in ref. 47 and applied in ref. 48
to describe atmospheric forcing of the extratropical oceans. This method has
been used successfully to explain the evolution of the PDO (9, 10, 14, 15),
although none of these prior studies have included the century-long trends in North Pacific circulation and SST. Monthly SLP1, SST1, SSTARC, and SATARC
indices are included in Dataset S1.
Monthly atmospheric and surface data obtained from 20CR (1900–2012)
include net turbulent heat fluxes (both latent and sensible), surface radiation
(solar and longwave, upward and downward), near-surface (10-m height)
winds, boundary-layer (2-m height) temperature, and specific humidity
(used to calculate relative humidity). Upper ocean dynamical and thermal
advective responses from SODA (1900–2008), which is forced with 20CR
surface winds, include near-surface (5-m depth) temperatures and horizontal
currents and subsurface temperatures and vertical velocities.
Linear regression was used to estimate monthly responses to a +1σ anomaly
of the leading pressure mode SLP1. Regressions were performed on monthly
indices over the 1920–2008 period during which both datasets are available
and 20CR circulation variability agrees well with the primary NCAR SLP
dataset (SI Text). Because century-long trends in 20CR heat fluxes and SODA
advection depart substantially from observed SST changes, regressions were
performed on indices treated with a 120-mo high-pass Hamming filter,
capturing the short-term responses reflected in the time series model. Statistical
significance was determined from Monte Carlo simulations using
random, identically filtered SLP1 series matching the observed level of
lag − 1 autocorrelation. Anomaly responses were computed separately
for each calendar month, and the overall responses were estimated as
the annual means of the monthly regression coefficients.
SATs around the NE Pacific margin were investigated with monthly station
data from the US Historical Climate Network, version 2 (USHCNv2) (49) and
the Global Historical Climate Network, version 3 (GHCNv3) (50), using adjusted
versions of both datasets. A NE Pacific Arc SAT index (SATARC) was
constructed from all 51 USHCNv2 and GHCNv3 coastal stations with at least
960 mo of available data (Fig. 1A). Monthly anomaly series were first calculated
for subregions of coastal Alaska (12 stations), British Columbia (3),
Washington-Oregon (19), California (14), and Hawaii (3). For each station,
missing data were infilled by monthly regression from up to three nearest
neighbors, defined by correlation coefficient. A small number of remaining
missing subregional estimates were filled by regression on neighboring subregions.
The final SATARC index was calculated as the unweighted mean of the
five subregional series, expressed in monthly degree Celcius anomalies with
respect to the full 1900–2012 period. Temperature indices were similarly
constructed for the US Pacific coast states, using all USHCNv2 stations to
produce monthly anomaly series for Washington, Oregon, and northern and
southern California, the latter divided at 37°N.

From the Results:
Leading Modes of NE Pacific SLP and SST
Empirical modes of NE Pacific (60°N–20°N, 180°W–100°W) SST
and SLP variability were initially determined by separate EOF
analyses of gridded normalized monthly anomalies for the 1900–
2012 period. The primary SST data come from the 2°-gridded
National Oceanic and Atmospheric Administration (NOAA)
Extended Reconstruction SST dataset, version 3b (ERSST)
(23). The leading SST mode (SST1) explains 30% of the total
monthly variance, primarily reflecting coherent SST variations
within an ∼1,000-km–wide band around the NE Pacific coastal
margins. This mode spatially resembles the PDO, but shows a
stronger relationship to SSTs near the North American coast and
a weaker connection to those in the central Pacific (Fig. 1A). The
monthly SST1 time series correlates most strongly with local SST
off northern California (r = 0.82 at 42°N, 128°W).
We formally define the NE Pacific Arc as the area within the
study domain where local monthly SST correlations with SST1
exceed 0.6 (Fig. 1A). The time series of SST1 expansion coefficients
is nearly identical (monthly r > 0.99) to an index of areaaveraged
SST anomalies within the Arc. This temperature index,
SSTARC, is used as a physically based metric of the statistical
EOF pattern. A 1 SD (σ) anomaly of the SST1 index corresponds
to an SSTARC deviation of 0.46 °C, with individual monthly relationships
deviating from this value by less than 10%. The monthly
SST1 index is positively correlated with the PDO index (monthly
r = 0.53), but displays considerably lower values before 1930,
partly due to the absence of the global SST trend in the PDO
index (Fig. 1B).
Primary SLP data come from the 5°-gridded National Center
for Atmospheric Research (NCAR) monthly analysis (24), previously
used in other studies of North Pacific low-frequency
variability (1, 2, 25, 26). The leading mode of NE Pacific SLP
(SLP1) explains 22% of overall variance, capturing coherent
anomalies over the central NE Pacific and most of the study domain
(Fig. 1C). The SLP1 index, oriented with positive values
reflecting lower pressure, correlates most strongly with SLP NE
of Hawaii (monthly r = −0.84) at 35°N, 150°W. These coordinates
lie near the center of the summertime North Pacific High, so that
SLP1 reflects variability of the subtropical anticyclone. In winter,
the center of the pattern lies between the climatological centers of
the Aleutian Low and North Pacific High, capturing southeastward
extensions of the cyclonic Aleutian Low circulation. The SLP1
index displays a positive long-term trend, signifying SLP reductions
over the NE Pacific from 1900 to 2012 (Fig. 1D). A comparison
with the inverted North Pacific Index (NPI) of Aleutian
SLP (27), developed from the same NCAR dataset indicates
stronger relative SLP changes in the east than in the center of the
North Pacific (Fig. 1D).