Record-low coastal sea levels in the Northeast Pacific during the winter of 2013–2014

During the winter of 2013–2014, the averaged tide gauge (TG) coastal sea level (CSL) anomaly north of 40°N was a record low of −107 mm for the period of 1948–2016. Statistical analysis indicates that this large drop was a once-in-a-century event and closely related to an unusual ocean warming event known as “The Blob”. The Blob developed in the NE Pacific during the winter of 2013–2014. Both the Blob and record-low CSL can be attributed to wind changes associated with an unusually high sea level pressure (SLP) pattern over the NE Pacific. The anomalous local longshore winds induced by the positive SLP anomalies caused strong offshore Ekman transport along the coast of NE Pacific, thereby leading to the record-low CSL. In addition, the steric sea level changes also contributed a significant part (17%) to the record-low CSL. The Pacific Decadal Oscillation (PDO), as the primary variability mode in the NE Pacific on decadal time scales, did not contribute to the emergence of this extreme CSL event.

Coastal sea levels (CSL) manifest important variability and changes of the coupled ocean-atmosphere system. Here we focus on the sea level change along the coast of the Northeast (NE) Pacific. In this region, CSL shows a near-zero or even decreasing trend over the past three decades, in sharp contrast to the global mean sea level rise [1][2][3][4] . Previous studies have indicated that the CSL trend in the NE Pacific is related to shifts in wind patterns associated with the Pacific Decadal Oscillation (PDO) [3][4][5][6] . While many coastal regions worldwide recently experienced record-high CSL, record-low CSL was reached in the NE Pacific during the winter of 2013-2014 (Fig. 1a, color), likely related to the so-called "Blob". The Blob refers to the unusually warm sea surface temperature (SST) in the NE Pacific in winter 2013-2014 with the largest anomalies of more than 2 °C (Fig. 1a, contour). Bond et al. 7 pointed out that the development of the extraordinarily warm SST anomalies in the NE Pacific can be attributed to anomalously weak surface winds induced by the strong positive sea level pressure (SLP) anomalies (Fig. 1b,  color), which suppressed the local ocean heat loss to the atmosphere and reduced horizontal advection and vertical mixing in the upper ocean. This extreme warm SST event and its effects on marine life have aroused widespread concerns [8][9][10] , as the NE Pacific ecosystems showed significant changes following the arrival of the Blob 9 . In contrast to ocean temperature, sea level signals associated with this significant event have not been investigated yet. Given that their relationship is established, sea level data could be used to quantify how extreme the Blob was in the NE Pacific.
The longshore component of the anomalous wind pattern induced by SLP anomalies is usually considered to be one of the main factors to influence CSL 6,11 . Sturges and Douglas 12 pointed out that longshore wind stress is one of the primary mechanisms for causing CSL change on the eastern side of the ocean and used longshore wind stress to discuss its effects on estimates of CSL rise. Bromirski et al. 4 showed that interdecadal trends in PDO-related wind stress curl can account for the suppression of CSL trends in the NE Pacific from 1980 to 2010. Moreover, they warned that an imminent phase shift in the PDO-related wind stress could lead to a sharp increase in NE Pacific CSL in the coming years. Thompson et al. 6 used the method of multiple linear regression to examine the relative contributions from remote equatorial wind stress, local longshore wind stress, and local wind stress curl, and suggested that the decreased rate of sea level rise along the coast of NE Pacific from 1992-2010 is primarily due to the ocean's response to the strengthened trade winds, with longshore wind stress becoming important to CSL trends from 37°N to 48°N. Most of these studies were concerned about the CSL trends in the NE Pacific over the past three decades, but extreme year-to-year changes have received less attention. The study  Table S1. VLM is estimated based on the difference between absolute sea level (ASL; AVISO) and relative sea level (RSL; TG) over the period of 1993-2016 15 . The decreasing trend of TG RSL is largely due to land uplift for the period of 1993-2016. In the present study, we estimate VLM based on the data after 1993. The trend of VLM may be different from 1948 to 2016. To gain insight into the cause of the 2013-2014 record-low CSL event in the NE Pacific, we mainly focus on the variables with the linear trends removed.
To further verify the emergence of the record-low CSL for the winter 2013-2014, we divide the coastal TG stations into two regions: north and south of 40°N, represented as TG_N and TG_S. The TG_N time series is calculated as the mean of six stations between Astoria and Yakutat and the TG_S time series is averaged of three stations for the south (San Diego, Los Angeles and San Francisco) (Fig. 2a,b). In addition to the detrended TG data, the detrended satellite data from AVISO and the detrended assimilation data from GODAS area-averaged over 1° × 1° boxes at the TG stations are also used to confirm the record-low CSL in the NE Pacific. Winter CSL anomalies for AVISO and GODAS are constructed by removing the winter mean computed over their corresponding period (AVISO in 1993-2016 and GODAS in 1980-2016). All three datasets show the lowest CSL anomalies during the winter of 2013-2014 (Fig. 2a). The most pronounced CSL drop associated with this event can be found in the TG data, in which the TG_N CSL anomaly decreased to −107 mm, much larger than that for AVISO (−52 mm) and GODAS (−68 mm). The difference of the three datasets can be due to the fact that the TG data is a point measurement at the coast while AVISO and GODAS gridded data are area averaged near the coast and outside the primary zone of Ekman transport convergence that peaks along the coastline. At the same time, we notice that the two maxima of winter CSL anomalies in Fig. 2a,b correspond to strong El Niño years (1982 and 1997), indicating the CSL anomaly in the NE Pacific may also reflect remotely-forced oceanic signals from the tropics 16 .
The 95% confidence intervals corresponding to the Gaussian distribution of the winter CSL anomalies over the period of 1948-2016 in the two regions are shown in Fig. 2c,d. For the winter of 2013-2014, the TG_N value is below the low limit of the confidence interval and the TG_S value is in the confidence interval. The 1982 and 1997 values are located above the upper limit of the confidence interval in both Fig. 2c,d, indicating CSL extremes during those two strong El Niño events. Except for the record-low value of winter 2013-2014, the value of winter 1955-1956 is also lower than the confidence interval, but not as extreme as that of 2013-2014. Furthermore, we Causes of the record-low CSL. The variability of CSL south of San Francisco is dominated by coastally trapped waves of tropical origin, while local wind stress is increasingly important to the north 6 . We find no record-low CSL value south of 40°N during winter 2013-2014 (Fig. 2b), suggesting that the signal of the extreme event is not from the tropics through coastally-trapped waves, but rather from the local wind stress forcing. Following Thompson et al. 6 , we use a multiple linear regression model to compare the contributions of local longshore wind stress (τ ls ), local wind stress curl (τ xy ) and remote equatorial wind stress (τ eq ) to the variability of TG_N winter CSL. τ eq is computed as the average zonal wind stress over the region spanning 6°N-6°S and 150°-280°E, τ xy is calculated area-averaged over 5° × 5° boxes at the TG_N stations. τ ls is computed by interpolating the vector wind stresses to the TG_N stations and projecting onto an approximate shoreline angle within a 2° radius of each TG_N station. The normalized time series of winter anomalies and the wind stress time series scaled by their respective regression coefficients are shown in Fig. S1. For the long-term time series, the result is consistent with Thompson et al. 6 that most of the variance of the regression results can be explained by τ eq compared to τ ls and τ xy . For the winter of 2013-2014, it can be mostly explained by the local longshore wind stress. The local longshore wind stress anomalies associated with the SLP anomaly in the NE Pacific (Fig. 1a) drives offshore Ekman transport, and thus lower CSL anomaly.
To understand the cause of the record-low CSL in the NE Pacific, the winter anomalies of SLP, upwelling index, CSL and SST spatially averaged for the TG_N stations are plotted in Fig. 3. The linear trends are also removed in each time series. During the winter of 2013-2014, the TG_N SLP anomaly is the highest during the analysis period (Fig. 3, red line). It drives an anticyclonic surface wind anomaly pattern over the NE Pacific. The local longshore wind stress anomalies produce offshore Ekman mass transport and cause strong coastal upwelling anomalies from southeast Alaska to northern California (Fig. 3, blue line). Therefore, the TG_N CSL reaches its lowest value during the winter of 2013-2014 (Fig. 3, black line). The correlation coefficients between TG_N CSL and the other three all reach their respective maximums at zero-lag and the correlation coefficient between CSL and SLP/upwelling index is significant (>0.6). We also find that the TG_N anomalies of SLP and upwelling index in the winter of 2013-2014 are also extreme values, further supporting the mechanism described above. www.nature.com/scientificreports www.nature.com/scientificreports/ In addition, the stronger upwelling (i.e., positive anomalies) also leads to negative SST anomalies (Fig. 3, blue line) and a decrease in seawater mass loading on the shelf along the west coast of North America. Stronger upwelling can bring cooler subsurface waters to the surface, thereby resulting in a negative steric sea level (SSL) anomaly due to seawater density changes. To evaluate the contribution of the SSL anomaly to the TG_N record-low CSL, we use gridded Argo temperature and salinity data to calculate the local SSL anomaly through equation (1) listed in the Methods section for the period of 2002-2016. Compared with the ERSST long term SST anomaly time series (Fig. 3, purple-solid line), the Argo winter SST anomaly time series (Fig. 3, purple-dashed line) depicts most of its variability. The calculated TG_N SSL anomaly (Fig. 3, black-dashed line) for the winter of 2013-2014 is −18 mm, which accounts for about 17% of the total sea level drop (−107 mm). Thus, the steric effect is an important factor. In the open ocean, the warmer seawater in the blob region could cause a 20 mm rise of SSL (Fig. S2d, red-dashed line), which explains most of the open ocean sea level anomaly (26 mm ; Fig. S2d, black line). This is consistent with the previous study by Wu et al. 17 , which found that in the northeast subtropical Pacific, the mass component is critical for the variability of CSL, while the steric effect dominates the open ocean sea level changes.

Role of large-scale climate modes. CSL changes have been linked to large-scale climate variations in
the NE Pacific such as the PDO 10,18,19 , North Pacific Gyre Oscillation (NPGO) 10,20 and El Niño 16,21 , and only the Kelvin waves associated with strong El Niño events typically affect CSL changes poleward of San Francisco 22 . For the winter of 2013-2014, the absence of a strong ENSO event precludes the possible role of coastal wave propagation in the extreme CSL event. We therefore focus on the role of PDO and NPGO. In Fig. S3, we show the normalized time series of winter PDO index (Fig. S3, red-dashed line), NPGO index (Fig. S3, blue-dashed line) and TG_N CSL anomaly (Fig. S3, black line), respectively. The correlation coefficient between TG_N CSL anomaly and PDO (0.48) is much larger than that between CSL anomaly and NPGO (−0.03), indicating that the variability of the winter CSL north of California is dominated by PDO rather than NPGO. This result is consistent with the studies of Di Lorenzo et al. 20 and Chhak et al. 23 , which pointed out the coastal upwelling and offshore volume/mass transport north of about 40°N is dominated by the PDO, whereas the NPGO dominates south of this latitude band along the California Current System. To estimate the correspondence between the NPGO/PDO and the TG_N record-low CSL, we use a multiple linear regression model to regress the anomaly time series of TG_N winter CSL onto winter PDO index and winter NPGO index. The results of the multiple linear regression and the regression coefficients are shown in Fig. 4a. For the winter of 2013-2014, the TG_N record-low CSL is −107 mm (Fig. 4a, black line), with slightly negative contributions from PDO and NPGO (1 mm; Fig. 4a, green line). The part that can be explained by the NPGO is negligible for the entire period (Fig. 4a, blue line), therefore confirming the TG_N CSL is not dominated by the NPGO. Similarly, the variability induced by the PDO is insignificant (−5 mm; Fig. 4a, red line), indicating the PDO did not play a role in this extreme event during 2013-2014. However, the PDO is more important in the winter of 2014-15 (Fig. 4a, red line). This is in line with the assessment by Di Lorenzo and Mantua 10 about the domination of PDO in the winter of 2014-15. This result is also consistent with the observed PDO index (Fig. 4b, black line) for the two winters.
It is worth noting that the observed PDO index in December 2013 is near zero, indicating that the winter of 2013-2014 is at the turning point of PDO from its negative (cold) to positive (warm) phase. To put this extreme event into context and further investigate the relationship between PDO and the CSL, we analyze the hindcast simulation of an eddy-resolving ocean model -LICOM. Figure 4b shows the comparison of the PDO index between observations (Fig. 4b, black line) and the model simulation (Fig. 4b, blue line). The correlation coefficient is significantly high (r = 0.90, P = 1.00). It indicates that the LICOM simulations can reproduce the large-scale climate modes reasonably well. The ensemble means of simulated sea level and wind stress anomalies for the cold phase shows negative CSL anomaly in the NE pacific (Fig. 4c, color), which is associated with anomalous southward longshore wind stress (Fig. 4c, vector). In the warm phase, we find positive CSL anomalies (Fig. 4d, color) and anomalous northward wind stress along the coast of NE Pacific (Fig. 4d, vector).
The winter of 2013-2014 is at the turning point of the PDO shift from its cold to warm phase, and from the lower to higher CSL in the NE Pacific. Although the PDO did not directly contribute to the emergence of this extreme event for the winter of 2013-2014, the subsequent shift of PDO to its warm phase (Fig. 4b, black line) can raise the 2014-2015 winter CSL in the NE Pacific (Fig. 4a, red line), thus making the CSL in the winter of 2013-2014 the lowest. Consistent with observations that show an intensification of the Aleutian Low in the NE Pacific during the winter of 2014-2015 (Fig. S4).

Discussion and Conclusions
We find that large-scale climate modes did not contribute to the emergence of the record-low CSL in the NE Pacific in winter 2013-2014. However, the NPGO, El Niño and PDO can be linked by the evolution of the positive SLP anomaly over the NE Pacific for the winter of 2013-2014 and the following two winters 10,[24][25][26][27] . The coastal SST anomaly is higher than normal in the winter of 2014-15 (Fig. 3, green line), corresponding to the 2014-15 NE Pacific marine heatwave 11 .
In the present study, we have investigated the winter CSL change in the NE Pacific during the period of 1948-2016 by using the TG, AVISO and GODAS datasets. We find that a record-low sea level event occurred during the winter of 2013-2014. The averaged TG CSL anomaly from southeast Alaska to northern California was −107 mm, which is a once-in-a-century event. The extreme event of CSL change is coincident with the occurrence of the Blob in the NE Pacific. Both phenomena are linked to a high atmospheric pressure system located near the NE Pacific coast. The higher-than-normal SLP led to an anomalous local longshore wind stress and thus caused the anomalous offshore Ekman transport. In addition, SSL change contributed about 17% of the total extreme. These ocean-atmospheric processes were responsible for the record-low CSL in the winter of 2013-2014, www.nature.com/scientificreports www.nature.com/scientificreports/ in sharp contrast to the near record-high globally averaged CSL during this same winter (https://www.climate. gov/news-features/understanding-climate/climate-change-global-sea-level).

Methods
Sea level datasets. We analyze three datasets to identify the variability of wintertime sea level along the coast of NE Pacific. Monthly mean tide gauge (TG) revised local reference dataset for 1948-2016 is obtained from the Permanent Service for Mean Sea Level (PSMSL) website (http://www.psmsl.org). The sea level measured by the tide gauge is a height to the level of benchmarks on the nearby land (Relative sea level; RSL), so vertical land motion will influence sea level change 28 . The inverted barometer (IB) effect on sea level is removed from the TG data using the NCEP/NCAR Reanalysis version 1 dataset. Second is the quarter-degree multiple altimeter data product, sea level anomaly, produced by Archiving, Validation, and Interpretation of Satellite Oceanographic (AVISO) covering the period from 1993 to 2016 (www.aviso.oceanobs.com). All classical geophysical and environmental corrections are applied, including the IB effect. Global mean sea level rise is removed for the period of 1993-2016. The sea level measured by AVISO is absolute sea level (ASL), vertical land motions (VLM) can be estimated based on the difference between ASL and RSL over the period of 1993-2016 15 . The third is dynamic sea level (sea surface height relative to geoid; DSL) from the NCEP Global Ocean Data Assimilation System (GODAS) datasets during the period of 1980-2016 (http://www.esrl.noaa.gov/psd/data/gridded/data.godas. html). This is ocean reanalysis data and does not include atmospheric pressure forcing, so the IB adjustment is not needed. Anomalies for the above datasets are constructed by removing the monthly climatology computed over their corresponding period (TG in 1948, GODAS in 1980and AVISO in 1993. Linear trends for the above datasets are removed over their corresponding period. Sea surface temperature (SST), sea level pressure (SLP) and wind stress datasets. To reconstruct the ocean and atmospheric variability associated with the NE Pacific winter anomalies we used NOAA extended reconstructed sea surface temperature (ERSST) V4 data, monthly mean SLP and wind stress data are obtained from NCEP/NCAR Reanalysis. We analyze the data over the period to 1948-2016. Anomalies for SST, SLP and wind stress are constructed by removing the monthly climatology over the period 1948-2016. The linear trends are removed from the anomalies at each grid point.