High‐Tide Floods and Storm Surges During Atmospheric Rivers on the US West Coast

Abstract Atmospheric rivers (ARs) cause inland hydrological impacts related to precipitation. However, little is known about coastal hazards associated with these events. We elucidate high‐tide floods (HTFs) and storm surges during ARs on the US West Coast during 1980–2016. HTFs and ARs cooccur more often than expected from chance. Between 10% and 63% of HTFs coincide with ARs on average, depending on location. However, interannual‐to‐decadal variations in HTFs are due more to tides and mean sea‐level changes than storminess variability. Only 2–15% of ARs coincide with HTFs, suggesting that ARs typically must cooccur with high tides or mean sea levels to cause HTFs. Storm surges during ARs reflect local wind, pressure, and precipitation forcing: meridional wind and barometric pressure are primary drivers, but precipitation makes secondary contributions. This study highlights the relevance of ARs to coastal impacts, clarifies the drivers of storm surge during ARs, and identifies future research directions.

. The frequency of HTFs along the US West Coast has increased in recent decades in some places (San Diego, La Jolla, San Francisco, and Seattle), and more generally shows interannual variability that correlates with phases of the El Niño-Southern Oscillation (ENSO; Sweet et al., 2021). However, few studies investigate the relationship between coastal sea level and ARs. Khouakhi and Villarini (2016) quantify the correspondence between ARs and extreme sea-level statistics on the US West Coast. They find that annual maxima of hourly still water levels at tide gauges between San Diego, California and Tofino, British Columbia occur within 12 hr of passing ARs 15-50% of the time. These authors also determine a relationship with modes of large-scale climate variability. For example, exceedances over the 99.5th percentile of the hourly still water level distribution during ARs occur more frequently during El Niños and less frequently during La Niñas. Shinoda et al. (2019) study the oceanic response to ARs during the CalWater 2015 field campaign. They observe daily averaged still water level anomalies of 30-50 cm at the Neah Bay, Washington and South Beach, Oregon tide gauges coinciding with landfalling ARs on January 16 and February 6, 2015. These authors determine that a high-resolution ocean general circulation model reproduces the timing of observed storm surges, but only about half of their magnitude. Shinoda et al. (2019) posit that the storm-surge response is mainly due to alongshore winds and coastal currents, and that model-data discrepancies reflect small-scale processes unresolved by the model.
These studies advance understanding of ARs and their impacts on sea level, but they also imply outstanding questions. First, the relationship between ARs and coastal impacts remains unclear. For instance, annual maxima and peaks-over-threshold statistics from Khouakhi and Villarini (2016) are not necessarily informative of HTFs. During years when no HTFs occur, annual maxima will not correspond to HTFs, whereas during years when multiple HTFs occur, some HTFs will not represent annual maxima. Likewise, the 99.5th percentile of a still water level distribution usually does not correspond to, and tends to be lower than, impact thresholds (Table S1 in Supporting Information S1; Sweet et al., 2018), meaning that many peaks-over-thresholds studied by Khouakhi and Villarini (2016) do not correspond to HTFs. Second, the factors driving storm surge during ARs remain to be established. For example, Shinoda et al. (2019) interpret storm surges during ARs in terms of the ocean's dynamic response to wind forcing. Their interpretation contrasts with Bromirski et al. (2017), who reason that the ocean's isostatic adjustment to barometric pressure is the primary mechanism of storm surge along the US West Coast. Khouakhi and Villarini (2016) recommend a future study to clarify the roles of wind and pressure forcing on storm surges during ARs.
Here we address these outstanding questions related to ARs, HTFs, and storm surges on the US West Coast. We consider tide-gauge data, HTF thresholds, a catalog of ARs, and a gridded atmospheric reanalysis to establish the relationship between ARs and HTFs as well as the factors forcing storm surge during ARs. Results suggest that ARs contribute significantly to HTFs on the US West Coast, and clarify the relative effects of wind, pressure, and precipitation forcing on the associated storm surges.

Data
We use hourly still water level observations, tidal predictions, and station datums for 24 tide gauges on the US West Coast from the National Oceanic and Atmospheric Administration (NOAA) Center for Operational Oceanographic Products and Services (CO-OPS). These records are selected because they are relatively long, complete, and span much of the US West Coast (Figure 1 and Table S1 in Supporting Information S1). They also represent the union of US stations considered either in past studies of ARs and sea level on the US West Coast (Khouakhi & Villarini, 2016;Shinoda et al., 2019) or in government reports on HTFs (e.g., Sweet et al., 2021), allowing us to interpret our results in light of past findings. location of an AR satisfying these criteria is defined as the reanalysis grid cell with the maximum IVT along the coast. The catalog includes the time, location, IWT, IVT, and zonal and meridional wind of ARs at their landfalling locations on a 2.5° × 2.5° grid along the US West Coast (22.5-57.5°N, 105-135°W; Figure 1) from January 1948 to March 2017. To complement information provided by the Gershunov et al. (2017) catalog, we also consider daily meridional and zonal wind stress, barometric pressure, and precipitation from the NCEP/NCAR Reanalysis 1. We also interrogated daily fields from the European Center for Medium-Range Weather Forecasts (ECMWF) Reanalysis Interim (Dee et al., 2011) and obtained comparable results (Figures S1 and S2 in Supporting Information S1).
We consider the data between January 1, 1980 and December 31, 2016. The start date is chosen partly based on the tide-gauge records, many of which begin in the late 1970s. By not considering data prior to 1980, we also avoid possible discontinuities in the reanalysis related to the advent of satellite data in the late 1970s. Data processing and methods specific to the analysis of either HTFs or storm surges are described in the next two sections before the respective results are introduced.

High-Tide Floods
We establish relationships between ARs and HTFs on the US West Coast using a peaks-over-threshold approach (cf. Khouakhi & Villarini, 2016). For each tide gauge, we count the number of days when HTFs occur for at least 1 hr (HTF days). We identify HTFs when still water levels exceed the local minor flood thresholds defined by Sweet et al. (2018), which range between 56 and 64 cm above local mean higher high water (Table S1 in Supporting Information S1). We also count the number of days when an AR passes nearby a tide gauge (AR days). An AR is nearby a tide gauge when it has IVT ≥ 500 kg m −1 s −1 at its landfalling location in the grid cell whose centroid is closest to the gauge ( Figure 1). Note that results are qualitatively insensitive to reasonable alternative definitions of "nearby" (Figures 3a and S7 in Supporting Information S1). We also count the number of days when both a HTF occurs and an AR passes nearby the gauge within ±24 hr of the HTF (HTF and AR days). Finally, we count the hypothetical number of days when HTFs would have occurred from mean sea-level changes and tides alone by first calculating and removing storm surge from the hourly water level data, and then identifying days when the flood threshold is exceeded. (The calculation of storm surge is detailed in the next section.) For all quantities of interest, we run 1,000 bootstrap iterations to estimate uncertainty due to the finite record length of the data (Text S1 in Supporting Information S1). We quantify statistical significance of the observed numbers of HTF and AR days and other cooccurrences by comparing observed values to values determined synthetically through 1,000 simulations of uncorrelated stochastic processes (Text S2 in Supporting Information S1).
HTF days and AR days along the US West Coast show clear spatial structure (Figures 2a and 2b). More HTF days and AR days were experienced on the Northwest Coast than the Southwest Coast. For example, San Diego, California experienced 79 ± 17 HTF days and 259 ± 30 AR days during the study period, whereas Neah Bay, Washington witnessed 329 ± 37 HTF days and 760 ± 54 AR days over that same time. All ± ranges identify 95% confidence intervals based on bootstrapping. The Puget Sound in Washington is an exception to the rule: fewer HTF days and AR days occurred at higher-latitude tide gauges in this estuary compared to lower-latitude tide gauges on the open-ocean coasts of Oregon and Washington, suggesting that these estuarine locations are more sheltered from processes driving HTFs and ARs. Central California also deviates from the trend, as fewer HTF days were observed at mid-latitude locations in this region compared to lower-latitude sites in Southern California. The basic patterns of HTF days and AR days found here are consistent with previous studies. For example, Sweet et al. (2021) report that more HTFs happen on the open coasts of Oregon and Washington than on the California coast or within the Puget Sound (their Appendix 1), while Neiman et al. (2008) find that more To clarify relationships between ARs and HTFs, we compute percentages of HTF days that are AR days and AR days that are HTF days (Figures 2c, 2d and 3a). The percentage of HTF days that are AR days quantifies whether ARs are a necessary condition for HTFs (values ∼100% indicate that HTFs only occur during ARs), while the percentage of AR days that are HTF days measures whether ARs are a sufficient condition for HTFs (values ∼100% indicate that ARs always lead to HTFs). On average along the coast, 28% ± 2.3% of HTF days are AR days, but values are elevated between Monterey and Arena Cove (48% ± 6.9%) in Central California, with the highest percentage (63% ± 19%) observed at San Francisco (Figures 2c and 3a). Similar findings were obtained using alternative criteria for evaluating if an AR is near a tide gauge (Figures 3a and S7 in Supporting Information S1), meaning that these results are robust to these subjective analysis choices. In comparison, since more AR days occurred than HTF days (Figures 2a and 2b), the percentage of AR days that are HTF days is lower on average (5.2% ± 0.4%), peaking more to the north, with 10% ± 1.1% of AR days being HTF days between Port Orford, Oregon and Toke Point, Washington (Figure 2d), suggesting that ARs alone are seldom sufficient to cause HTFs. Nevertheless, at nearly all sites, values in Figures 2c, 2d and 3a are statistically significant (p < 0.05), meaning that HTFs and ARs cooccur more often than expected from random chance, and suggesting that ARs are important contributors to HTFs.
HTF and AR frequencies also vary across time (Figure 3b). The annual number of HTFs averaged along the US West Coast varies from 0.7 ± 0.7 to 13 ± 5.9 days per year, while the average number of ARs ranges between 7.2 ± 3.1 and 21 ± 6.3 days per year (Figure 3b). HTF days were highest in 1982 (13 ± 5.9 days) and 1997 (12 ± 5.4 days) during strong positive ENSO events. This observation is consistent with past studies identifying a relationship between ENSO and HTF frequency on the US West Coast (Sweet & Park, 2014;Sweet et al., 2021). The Pearson correlation coefficient between interannual variations in HTF and AR days on the US West Coast (0.2 ± 0.2) is not statistically significant (p > 0.05). In contrast, the number of HTF days per year is significantly correlated with annual mean sea-level anomaly averaged along the coast (0.7 ± 0.1, p < 0.01; Figure 3b). An even Figure 2. Number of (a) high-tide flood (HTF) days and (b) atmospheric river (AR) days at tide gauges during 1980-2016. Percentage of (c) HTF days experiencing ARs, and (d) AR days experiencing HTFs. The "x" at Santa Monica, California in panels (c) and (d) indicates that the value is not significant given the null hypothesis of two uncorrelated stochastic Poisson processes (Text S2 in Supporting Information S1).
higher correlation (0.9 ± 0.1, p < 0.01) is found between observed HTF days and hypothetical HTF days expected from tides and mean sea-level changes, such that the latter explains 66 ± 14% of the variance in the former, suggesting that changes in these extreme sea-level events are governed more by tides and mean sea-level changes than changes in storminess (cf. Marcos et al., 2015;Menéndez & Woodworth, 2010;Ray & Merrifield, 2019;Thompson et al., 2021).

Storm Surges
We quantify storm surges and their causes during ARs on the US West Coast using a composite analysis (cf. Shinoda et al., 2019). We identify all ARs passing by tide gauges during the study period. For each AR as it passes by a gauge, we isolate the day when maximum IVT takes place and interpret it as when the gauge experiences the strongest effect of the AR. We then take the associated daily storm surge from the tide gauge, which we calculate from daily mean still water level by removing the tidal prediction, seasonal cycle, and linear trend, and then applying a high-pass filter based on a 20-day moving-median operator. We use the 20-day time scale to separate higher-frequency storm surges from lower-frequency changes in mean sea level, similar to past studies (Dangendorf et al., 2016;Serafin et al., 2017). The moving median represents a robust tool for detecting extremes in the presence of trends and other noise (Mudelsee, 2020).
Storm surges during ARs show clear spatial structure (Figures 4a, 4b and 5a). Surges are larger on average at higher latitudes (Figures 4a and 5a). Mean storm surge during an AR grows from 3.1 ± 1.2 cm at Santa Monica, identify different criteria applied to determine whether an AR is nearby during a HTF (i.e., whether the minimum IVT threshold is 250 or 500 kg m −1 s −1 and 1 or 2 nearby grid cells are considered). Black is the null hypothesis for two random stochastic Poisson processes (IVT ≥ 500 kg m −1 s −1 nearest 1 grid cell; Text S2 in Supporting Information S1). (b) Averages across all tide gauges along the US West Coast of yearly observed HTF days (blue), AR days (orange), and annual mean sea level (black). Thick lines and shaded values are, respectively, bootstrap estimates of average values and 95% confidence intervals. Blue dashed line is the best estimate of the number of HTF days per year expected hypothetically from tides and mean sea-level changes (see text for details). Note that the horizontal axis has units of meteorological years (May-April).
California to 21 ± 3.2 cm at Toke Point, Washington. Deviations from this trend are apparent at locations in the Puget Sound, where mean surge values are lower than expected from latitude alone, which could reflect important estuarine processes distinct from the mechanisms that mediate storm surge along the open-ocean coastline. Storm surge is also more variable at higher latitudes (Figure 4b). For example, the standard deviation of storm surge during ARs is 4.3 ± 0.8 cm at La Jolla, California, 12 ± 1.6 cm at South Beach, Oregon, and 20 ± 5.3 cm at Toke Point, Washington. [Note that, while we use mean and standard deviation as summary statistics, storm surge distributions are not Gaussian ( Figure S3 in Supporting Information S1).] Such surges are rarely large enough, when superimposed on mean higher high water, to overtop flood thresholds (cf. Table S1 and Figure S3 in Supporting Information S1). This corroborates the suggestion made in the previous section that ARs alone are seldom sufficient to cause HTFs.
These basic patterns are qualitatively consistent with previous numerical studies of sea level and ARs as well as past observational studies of storm surge in the region. Considering tide-gauge data during 1935-2014, Bromirski et al. (2017 show that the 99th percentile of hourly nontidal winter residuals increases steadily from 10 to 15 cm in Southern California to 45-55 cm in Oregon and Washington (their Figure 2c). Serafin et al. (2017) reveal that the average and spread of observed annual maxima in hourly nontidal residuals from 11 tide gauges between La Jolla, California and Neah Bay, Washington, increase from south to north along the coast (their Figure  1e). Using a high-resolution ocean general circulation model, Shinoda et al. (2019) report that coastal sea level rises during the days leading up to an AR by between ≲1 cm off Southern California to ≳4 cm off Oregon and Washington (their Figure 8h). However, these studies do not establish what processes drive storm surge during landfalling ARs.
To attribute observed surges (Figures 4a and 4b), we use contemporaneous daily zonal and meridional wind stress, barometric pressure, and precipitation from NCEP/NCAR Reanalysis 1 at the grid cells closest to the tide gauges. We remove seasonal cycles and linear trends from the reanalysis and apply a 20-day high-pass filter. To quantify how much storm surge can be understood in terms of local wind, pressure, and precipitation anomalies, we consider a simple model that represents surge as a linear superposition of the atmospheric forcing  (1) reflects physical considerations. On these time scales, we anticipate a transient adjustment, such that the ocean's response may be out of phase with the atmospheric forcing (Gill, 1982). The Hilbert transform, or quadrature function, rotates each Fourier component of a time series by ±90° (Thomson & Emery, 2014). Therefore, by including forcing terms and their Hilbert transforms in Equation 1, we represent arbitrary phase relationships between forcing and response. In contrast, a regression model that included the forcing terms but omitted their Hilbert transforms would only permit in-phase or antiphase relationships. For clarity, let ̂ , ̂ , ̂ , and ̂ identify the modeled ζ responses to π, τ, p, and q forcing, respectively, and ̂ the total modeled ζ response. We use ridge regression to determine the a's and b's at each tide gauge (Text S3 in Supporting Information S1), which is preferable to ordinary least squares given possible collinearity between predictors. Results are based on a ridge-parameter value of λ = 0.3, but similar findings follow from a range of λ values ( Figure S4 in Supporting Information S1). Modeled ̂ shows skill in explaining ζ observed at tide gauges (Figures 4 and 5). The model reproduces the observed structure that surges grow larger and more variable with latitude along the coast (Figure 4). Mean storm surges from the observations ζ and the model ̂ overlap within estimated uncertainties everywhere on the California coast (Figure 5a). Along Oregon and Washington, the model can underestimate observed mean storm surge (by as much as 32% on average at Cherry Point, Washington), possibly due to shrinkage related to the ridge regression, reanalysis errors (e.g., due to coarse grid cells that overlap land and sea), or processes absent from the model (Figure 5a). The model also accounts for most of the observed storm-surge variation at all gauges (Figure 5b), explaining between 57 ± 20% (La Jolla, California) and 87 ± 3.4% (Point Reyes, California) of the variance in the data, where the percent variance  in a variable x explained by another variable y is defined as and var is the variance operator. As defined in Equation (2),  is similar to the familiar Nash-Sutcliffe model efficiency coefficient from hydrology (Gupta et al., 2009;Li, 2016;Nash & Sutcliffe, 1970).
The model is also informative of the relative influences of π, τ, p, and q forcing on ζ ( Figure 5). Primary contributions to ζ are made by p and τ ( Figure 5). On average, ̂ contributions to mean ζ values are nearly spatially uniform along the coast, ranging between 2 and 5 cm (Figure 5a). In contrast, average ̂ values become larger with latitude, growing from 0.3 ± 0.9 cm at Santa Monica, California to 11 ± 2.3 cm at Toke Point, Washington. In Southern California and within Puget Sound, ̂ is the more important contributor to ζ variance, but elsewhere ̂ and ̂ contribute comparably (Figure 5b). Forcing by q can also make secondary contributions ( Figure 5). Mean ̂ values are distinguishable from zero at most sites, reaching as high as 2.3 ± 0.8 cm in Point Reyes, California and 3.2 ± 2.0 cm in Toke Point, Washington (Figure 5a). In and around San Francisco Bay, and along portions of the Washington coast, ̂ explains 10-20% of the ζ variance on average (Figure 5b). In contrast, π forcing is largely insignificant ( Figure 5). In most places, estimates of ζ variance explained by ̂ overlap with zero (Figure 5b), and mean ̂ values are indistinguishable from zero or small and negative (Figure 5a). Two reasons may together explain our finding that π is not an important contributor. First, τ anomalies, which are mostly alongshore on the US West Coast, are typically stronger than π anomalies, which are mainly onshore, during ARs at gauges ( Figure  S6 in Supporting Information S1). Second, ζ can be more sensitive to an alongshore wind stress anomaly than to an onshore wind stress anomaly of equal magnitude (Text S4 and Figures S4, S5 in Supporting Information S1). More generally, spatial structures apparent in storm-surge contributors ( Figure 5) may have partly to do with variations in the orientation or strength of ARs along the coast.

Summary and Discussion
Atmospheric rivers (ARs) typically bring heavy rain, strong wind, and low pressure to the coastal zone. We established relationships between ARs and high-tide floods (HTFs), and identified forcing mechanisms responsible for storm surge during ARs on the US West Coast during 1980-2016. ARs and HTFs cooccur more often than expected from random chance, and 10-63% of HTFs coincide with ARs, depending on location (Figures 2 and 3). Interannual variations in HTF days and AR days per year are not significantly correlated (Figure 3), meaning that more ARs do not necessarily result in more HTFs. Instead, there is a significant correlation between observed HTF days per year and the HTF days expected from tides and mean sea-level changes alone (Figure 3). A linear model including local wind, pressure, and precipitation forcing accounts for ≥68% of the average magnitude and 57-87% of the variance in magnitude of storm surges during ARs (Figures 4 and 5). Meridional wind and barometric pressure make primary contributions to storm surge, but precipitation has a secondary effect in some places ( Figure 5).
HTFs arise from a subtle interplay of distinct processes acting on different timescales. While they tend to occur and be most acute at high tide, HTFs are generally not caused by tides alone (cf. Hague & Taylor, 2021). Fewer HTFs would occur on average from tides and mean sea-level changes in the absence of surges due to ARs and other events (Figure 3). However, surges associated with ARs are rarely large enough, when added to mean higher high water, to cause HTFs on their own (Figure 3). It is only when superimposed on a favorable tide or mean sea-level anomaly that storm surges related to ARs are generally capable of exceeding HTF thresholds. For a full understanding of observed HTFs, the effects of surges, tides, and mean sea level must all be considered.
This paper advances knowledge of hazards related to ARs and the oceanic response to atmospheric forcing on the US West Coast. Past studies emphasize hydrological impacts of ARs related to extreme precipitation , but we show that ARs also drive coastal impacts related to sea level. By quantifying relationships between HTFs and ARs, and identifying the factors driving storm surge during these events, we resolve outstanding questions in the literature (Bromirski et al., 2017;Khouakhi & Villarini, 2016;Shinoda et al., 2019). This paper sheds light on HTFs, occurrences of which are increasing on much of the US Coast , and will accelerate into the future (Thompson et al., 2021). Our work is consistent with the notion that observed changes in sea-level extremes are attributable more to changes in mean sea level and the tides than to changes in storminess (Marcos et al., 2015;Menéndez & Woodworth, 2010;Ray & Merrifield, 2019;Thompson et al., 2021). Our results also underscore the importance of understanding locally forced high-frequency sea-level variability on the US West Coast (Battisti & Hickey, 1984;Bromirski et al., 2017;Chapman, 1987;Gill & Clarke, 1974;Ryan & Noble, 2006;Verdy et al., 2014).
We conclude with some limitations of our study and future research directions.
1. Space constraints precluded a complete study of the spatiotemporal statistics of HTFs and ARs on the US West Coast. Future studies should consider more granular details, such as temporal variation in HTF and AR cooccurrences at individual tide gauges across various timescales, including the seasonal cycle and decadal trends, to identify whether sea-level rise influences the covariance between HTFs and ARs, and if HTFs due to ARs occur mainly in particular months of the year (Thompson et al., 2021). 2. We focused on the US West Coast, but ARs make landfall in other midlatitude and high-latitude regions . Links should be established between ARs and sea-level extremes on a more global basis (cf. Carvajal et al., 2021;Ridder et al., 2018). 3. We used the AR catalog of Gershunov et al. (2017), but other AR data sets are available, which differ in terms of their detection algorithms Shields et al., 2018). These differences can affect the frequency, intensity, and duration ARs identified by the data sets (Ralph et al., 2019). For example, considering a catalog using more selective criteria that recognized fewer, more intense AR events, we may find higher average storm surge (cf. Figures 4 and 5) but smaller percentage of HTF days that are AR days (cf. Figures 2 and 3). A thorough error analysis is beyond our scope, but we considered 22 other AR data sets participating in the "Tier 1" Atmospheric River Tracking Method Intercomparison Project (ARTMIP; Text S5 in Supporting Information S1). Findings based on the Gershunov et al. (2017) catalog are consistent with and representative of results obtained from the other ARTMIP data sets more generally ( Figure S7 in Supporting Information S1). This suggests that, had we used another AR catalog for our main analysis, we would expect similar results on average. However, uncertainties on the estimates from the various ARTMIP data sets can be large ( Figure S7 in Supporting Information S1). A future study should identify the origins of these uncertainties and what AR catalogs are most informative for studies of coastal impacts. 4. We focused on storm surge and HTFs, but ARs could affect other quantities of interest to coastal impacts, such as waves and erosion (Serafin et al., 2017;Theuerkauf et al., 2014). A more comprehensive assessment of coastal hazards due to landfalling ARs, including their role in compound events (AghaKouchak et al., 2020), should be made. 5. We used flood thresholds from the common impact threshold framework of Sweet et al. (2018), which is a consistent national coastal flood metric, applicable everywhere tidal datums are established. However, flood thresholds based on this framework may be lower or higher than levels that correspond to local impacts (Kriebel & Geiman, 2013). The sensitivity of our results to other definitions of flood threshold should be quantified. 6. Our investigation of storm surge was statistical in nature. Regression coefficients found empirically from the data are consistent with basic expectations for a coastal-trapped barotropic sea-level response to local wind, pressure, and precipitation forcing over a frictional shelf (Text S4 and Figure S4 in Supporting Information S1), suggesting that we identify causal relationships between storm surge and atmospheric forcing. Even so, a more physics-based assessment would be informative, allowing the relative roles of the various (correlated) forcing mechanisms to be more unambiguously identified. Because precipitation is not often identified as a driver of storm surge (Gill, 1982;Pugh & Woodworth, 2014), it would be particularly informative to test our hypothesis that precipitation can contribute to storm surge during ARs. 7. We used observations of the past four decades, but the nature of ARs could change under future warming. While their dynamical response to climate change remains uncertain (Shepherd et al., 2014;Vallis et al., 2015), ARs are expected to become more frequent (Espinoza et al., 2018), contain more moisture (Dettinger, 2011), and shift poleward (Yin, 2005) as the climate changes. It remains to evaluate how future changes in ARs would aggravate coastal impacts already expected from future sea-level rise (Jevrejeva et al., 2019;Kopp et al., 2017).