Climatic Drivers of Extreme Sea Level Events Along the Coastline of Western Australia

Accurate prediction of coastal flooding requires a detailed understanding of all individual contributions to sea level variability and how they interact to trigger extreme sea level (ESL) events. In this study, we focus on the expansive (∼10,000 km) coastline of Western Australia, a region that experiences large latitudinal gradients in met‐ocean sources of sea level variability, as a case study to investigate the mechanisms responsible for ESLs and trends over the past 54 years (1966–2019). Using long‐term sea level records from tide gauges and satellite altimetry, we explore how different contributions to sea level variability at different time scales (from hourly to interannual) interact to generate ESLs. We observe that all individual, nontidal contributions to ESLs (i.e., atmospheric surge, seasonal and interannual variability) are of similar magnitude (of order 10 cm) along the entire coast and comparable to the tidal variations in the microtidal southwestern region. The results reveal the important role that seasonal and interannual sea level variability plays in generating ESLs, with these low‐frequency contributions being relatively large compared to typical global values. With mean sea level having risen by ∼10 cm over this 54‐year study period, sea level rise was also identified as making an increasingly significant contribution to observed increases in the frequency of ESLs. Overall, due to the comparatively large magnitude of low‐frequency sea level contributions (seasonal and longer), the Western Australia coast provides a useful case study to investigate how sustained periods of elevated sea level will impact coastlines worldwide more broadly in the future.


Introduction
Extreme sea level (ESL) events are a primary driver of coastal flooding along most coastlines worldwide (Rueda et al., 2017;Ruggiero et al., 2001;Vousdoukas et al., 2018;Wahl et al., 2017). An ESL event is usually triggered by positive interactions between multiple individual components of sea level variability that occur over a wide range of time scales (from seconds to decades and longer), including tides, atmospheric surges, seasonal and interannual sources of water level variability, wave runup, and long-term mean sea level trends (Melet et al., 2018;Merrifield et al., 2013;Serafin et al., 2017;Woodworth et al., 2019). The superposition of mean sea level rise (SLR), which has averaged ∼3 mm year −1 globally in recent decades (but even exceeding 10 mm year −1 in some regions) (Church et al., 2013) makes ESL events more likely, with their frequency and magnitude increasing at varying rates worldwide Vousdoukas et al., 2018). These Abstract Accurate prediction of coastal flooding requires a detailed understanding of all individual contributions to sea level variability and how they interact to trigger extreme sea level (ESL) events. In this study, we focus on the expansive (∼10,000 km) coastline of Western Australia, a region that experiences large latitudinal gradients in met-ocean sources of sea level variability, as a case study to investigate the mechanisms responsible for ESLs and trends over the past 54 years . Using long-term sea level records from tide gauges and satellite altimetry, we explore how different contributions to sea level variability at different time scales (from hourly to interannual) interact to generate ESLs. We observe that all individual, nontidal contributions to ESLs (i.e., atmospheric surge, seasonal and interannual variability) are of similar magnitude (of order 10 cm) along the entire coast and comparable to the tidal variations in the microtidal southwestern region. The results reveal the important role that seasonal and interannual sea level variability plays in generating ESLs, with these low-frequency contributions being relatively large compared to typical global values. With mean sea level having risen by ∼10 cm over this 54-year study period, sea level rise was also identified as making an increasingly significant contribution to observed increases in the frequency of ESLs. Overall, due to the comparatively large magnitude of lowfrequency sea level contributions (seasonal and longer), the Western Australia coast provides a useful case study to investigate how sustained periods of elevated sea level will impact coastlines worldwide more broadly in the future.
Plain Language Summary Accurate prediction of coastal flooding requires a detailed understanding of the diverse contributions to sea level variability and how they interact to trigger extreme sea level events. In this study, we focus on the expansive (∼10,000 km) coastline of Western Australia, a region that experiences large differences in the sources of sea level variability, as a case study to investigate the causes of sea level extremes over the past 50 years. We combine coastal tide gauge observations and satellite sea level measurements to explore how different contributions to sea level over different time scales (ranging from hours to decades) generate extreme events, including how these contributions have evolved over time. The results reveal the important role that seasonal and interannual sea level variability plays in generating extreme events, which are relatively large compared to typical global values. With mean sea level having risen by ∼10 cm over the study period, sea level rise was also identified as making an increasingly significant contribution to the elevated frequency of extreme events. The Western Australia coast provides a useful global case study to investigate how sustained periods of elevated sea level will impact coastlines more broadly in the future. LOWE ET AL. regional variations in the atmospheric and oceanic drivers of individual sea level contributions can lead to substantial spatial and temporal variability in sea level that must be accounted for when forecasting ESLs for a specific coastal region (Menéndez & Woodworth, 2010). Therefore, to develop robust predictions of the probability of ESL events for a given coastal region first requires establishing a detailed understanding of the role that different sea level contributions play in driving ESLs (Losada et al., 2013;Marcos & Woodworth, 2017;Méndez et al., 2007;Serafin & Ruggiero, 2014).
The expansive (∼10,000 km long) coastline of Western Australia (WA) (Figure 1) experiences diverse sources of sea level variability along a latitudinal gradient due to regional variability in its dominant metocean drivers (McInnes et al., 2016;Pattiaratchi & Eliot, 2009). The northwest coast experiences a large semi-diurnal tidal regime (range exceeding 8 m in some coastal regions), whereas the southwest coast experiences microtidal diurnal tidal conditions (mean range only ∼0.5 m) (Eliot, 2012). These tidal ranges are also modulated at long (interannual) time scales by an 18.6-year lunar nodal cycle (more important in the southwest) and the 4.4-year lunar perigean subharmonic cycle (more important in the northwest). The influence of the 18.6 lunar nodal cycle can be especially important in the microtidal southwest where it can modulate tidal ranges by up to ∼10 cm, equivalent to 20%-30% of the mean tidal range (Eliot, 2010). The tropical northwest is also one of the most active tropical cyclone regions in the world, experiencing an average ∼5 cyclones each year (Goebbert & Leslie, 2010), which can episodically generate large atmospheric surges during the austral summer months and transmit these along the coast to the south as coastally trapped waves. Similarly, the southwest coast can also experience large surges during extratropical storms that are common in winter months (Eliot & Pattiaratchi, 2010). At longer time scales (seasonal to interannual), sea level variability along WA is strongly influenced by its unique poleward-flowing eastern boundary current system (the Leeuwin Current) that generates seasonal sea level fluctuations of tens of centimeters via geostrophic balances along the shelf, with a peak in sea level during the months of March-June (Feng et al., 2003;Ridgway & Godfrey, 2015). The strength of the Leeuwin Current can also vary strongly over interannual time scales due to the influence of the El Niño-Southern Oscillation (ENSO) cycle that transmits sea level anomalies between the Pacific and Indian Oceans (Merrifield et al., 2012;Wijffels & Meyers, 2004).
A number of global studies have mapped the relative importance of individual sea level contributions that contribute to ESLs around the world and have identified how ESL statistics have evolved over the past century (e.g., Losada et al., 2013;Menéndez & Woodworth, 2010;Merrifield et al., 2013). However, further work is needed to understand how individual sea level contributions will interact over time (on an event-by-event basis) to generate ESLs, which will importantly require improving understanding of the underlying regional oceanic and atmospheric processes that are responsible for driving ESLs now and in the future. The diverse sources, patterns, and time scales (including phasing) of sea level variability along the WA coastline, combined with the background long-term SLR trends, serve as a useful global case study to investigate how different atmospheric and oceanic processes interact to drive ESLs along a single continuous coastline.
In this study, we analyze long-term tide gauge records over the past 54 years (1966-2019) and satellite altimetry observations (recent 26 years) to investigate how different oceanic and atmospheric processes interact to trigger ESL events along the WA coastline. Using these observations, we assess how the frequency of ESL events has been changing over the past 54 years and isolate the mechanisms responsible for these trends. Through our analysis we identify how global climate cycles trigger interannual clustering of ESL events and how regional SLR is increasing the probability of extremes. We discuss implications for how the results can be used to help understand and forecast ESLs over different time scales. LOWE ET AL. 10.1029/2020EF001620 2 of 16

Data Sources
Historical sea level records were obtained at seven locations along the WA coastline based on tide gauge observations available from the WA Department of Transport (https://catalogue.data.wa.gov.au/dataset/ tide-stations) (Figure 1). Although there are several additional tide gauges along the WA coastline, the seven sites analyzed here were chosen because they each contained a near-continuous record of hourly sea levels over the past 54 years between 1966 and 2019, with each site containing between 84% and 99% valid data over the study period. Sea levels from each tide gauge (available at an hourly interval) were adjusted from lowest astronomical tide to the Australian Height Datum (AHD), defined based on the mean sea level over the period 1966-1968 (hence coinciding with roughly the start of the analysis period), using the offset from an adjacent survey benchmark. We note that using the historical tide gauge data here, we are strictly assessing relative sea level changes; however, in Section 3.1 and supporting information, we also consider absolute sea level trends, as estimated by satellite altimetry and find comparable agreement over the study period (see also White et al., 2014). By assessing sea level variations using tide gauge data, we focus specifically on identifying the mechanisms responsible for variations in total sea level (still water level) and thus do not consider potential wave-driven sources of coastal flooding (i.e., contributions from wave runup) in this study.
Regional patterns in annual (seasonal) and interannual sea level variability in the south-eastern Indian Ocean surrounding WA were analyzed using gridded monthly averaged satellite altimetry data at 0.25° resolution available from SSALTO/DUACS (available at https://www.aviso.altimetry.fr/en/data) for the period 1993-2016. Relationships between interannual sea level variability and climate state (i.e., ENSO) were evaluated based on the Southern Oscillation Index (SOI) and the Multivariate ENSO Index (MEI) (available at https://www.esrl.noaa.gov/psd/enso/). Other climate indices were also initially considered (e.g., the Dipole Mode Index associated with the strength of the Indian Ocean Dipole) but were found to be weakly correlated with interannual sea level variability and are thus not discussed here. Finally, atmospheric forcing variability (i.e., surface air pressure) was obtained over the period 1979-2019 using the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis, available hourly on a 30-km grid (available at https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era5), with data extracted from each grid cell nearest each tide gauge.

Decomposition of Sea Level Variability
The hourly sea level time-series η(t) as a function of time t at each tide gauge was decomposed into three fundamental contributions as where η MSL is the mean sea level trend over the record, η T is the astronomical tide contribution, and η NTR is the nontidal residual. A number of approaches have been proposed to decompose the contributions of coastal sea level variability (e.g., Melet et al., 2018;Merrifield et al., 2013), which are generally similar, but can use subtly different definitions and analysis approaches. In this study, we adopt a similar approach to Serafin et al. (2017). The mean sea level trend contribution (η MSL ) was determined from a linear regression of the annual-averaged sea levels over each record. To determine the contribution of the tide (η T ), a tidal harmonic analysis using UTIDE (Codiga, 2011) was applied using 63 constituents, with the annual (S a ) and semiannual (S sa ) tidal constituents removed since the astronomical contribution of these long-period constituents is negligible (order 1 mm at these latitudes) (Lisitzin, 1974) relative to the much larger (order 10 cm) annual contribution of nontidal seasonal sea level variations forced by large-scale atmospheric and oceanic processes (discussed further below). The nontidal residual sea levels (η NTR ) were thus obtained by subtracting the mean sea level trend (η MSL ) and tidal contribution (η T ) from the measured sea level time-series (η).
The nontidal residual sea level time-series (η NTR ) was then further decomposed as where η A is the seasonal (annual) contribution, η I is the interannual contribution, and η SS is the high-frequency "storm surge" contribution. The seasonal contribution η A , including the amplitude and phase, was obtained by fitting annual (12 months) and semiannual (6 months) harmonics to the η NTR time-series using a least squares approach, with the semiannual harmonic included to account for some nonlinearity of the seasonal cycle (Merrifield et al., 2013). The interannual variability (η I ) was defined by subtracting the seasonal contribution (η A ) from η NTR and then monthly averaging the resulting data to remove daily to weekly variations (e.g., due to atmospheric surges) (Serafin et al., 2017).
The "storm surge" contribution (η SS ) was defined as the remaining high-frequency (i.e., daily to weekly) contributions to the nontidal residual sea levels, which are commonly associated with atmospheric storm surges at weather band frequencies of order day to week. To reduce the potential for some tidal contamination that is not completely removed in the η NTR signal, a 40-h moving average filter using a Hamming weighting function was applied to the η SS time-series. We note that while η SS will be referred to as a storm surge contribution based on the frequencies of this variability (as is commonly done), it is important to acknowledge that other high-frequency sources of sea level variability (order weekly or less) would be incorporated in this contribution; for example, due coastally trapped waves and high-frequency variability in the Leeuwin Current (including encroaching eddies) (Woodworth et al., 2019).
Local atmospheric contributions to storm surges are, in general, influenced by surface air pressure fluctuations (i.e., the "inverse barometer effect") and wind stress effects (wind setup). While the static sea level response to atmospheric pressure changes (inverse barometer effect) can be readily estimated from the hydrostatic pressure adjustment as (Ponte, 1994), where P atm is atmospheric pressure (with the overbar denoting the long-term mean), ρ is the seawater density, and g is gravity, the response to wind stresses is often much more complicated due to how it depends on coastline orientation to winds, tidal currents, Coriolis, local bathymetry, and other factors (Bode & Hardy, 1997). Given that a detailed investigation of the dynamics of the local storm surge responses is beyond the scope of this study, we instead adopt a simple approach that explores the links between η SS and the local atmospheric pressure variability, given that atmospheric pressure and wind variability tend to be closely related. For consistency with the definition of η SS , high-frequency fluctuations (e.g., diurnal and shorter) were filtered from the hourly pressure time-series by applying the same 40-h moving average filter (Hamming weighting function). A lagged correlation analysis between atmospheric pressure and η SS was then applied at each site, with significance levels computed based on the effective degrees of freedom computed from the record length divided by the autocorrelation time scale (Emery & Thomson, 2001). In addition, the dominant time scales of atmospheric pressure and surge variability were assessed by computing variance-preserving spectra    2 f (frequency times power spectral density), expressing how signal variance is distributed in logarithmic frequency space (Emery & Thomson, 2001).
Finally, the hourly reconstructed sea level time-series (sum of all separate contributions, SS ) were compared to the original sea level records to confirm that the approach accurately captured the observed sea level variability, including extreme events. The residual error for all sites, defined based on computing the Normalized Root Mean Squared Error (normalized by the local range of the water level records) ranged between 1% and 3% across all sites (Table 1).

Extreme Event Analysis
ESL events were defined as the top 2% of the hourly sea level values within each tide gauge record that were separated by at least 3 days to avoid duplicate counting of individual events. For each extreme event, the instantaneous breakdown of the individual sea level contributions that were responsible for the event were recorded. Over the 54-year period, the annual frequency of extreme events was evaluated, as well as the mean contribution of individual sea level contributions to the extreme events occurring each year.
To assess how the probability of extreme events varied over the record, a nonstationary extreme value analysis was used based on a Generalized Extreme Value (GEV) distribution applied to annual maxima values: is the scale parameter, and    c x is the shape parameter, which are (in general) allowed to vary as a function of a covariate x c (e.g., time). By comparison, in a conventional stationary extreme value analysis, the parameters ,  , and  are treated as constants, which assumes no changes in the statistical properties that govern the frequency of extremes over time. The analysis was specifically implemented using the Process-Informed Nonstationary Extreme Value Analysis (ProNEVA) tool, which estimates the parameters (including uncertainties) using a Bayesian approach with Markov Chain Monte Carlo sampling (refer to Ragno et al., 2019, for details). While it is possible to account for the nonstationary behavior by assuming various functional relationships with x c for the location, scale, and shape parameters, we chose to assess simple linear functions of these parameters with covariate x c given the length of the data records. Given that the analysis was based on block (annual) values, we focused on assessing how longer-term (>1 year) sources of variability in the records influenced exceedance levels and return intervals, focusing specifically on time (t) and climate indices (i.e., SOI) as the covariates x c . Therefore, the analysis first focused on investigating how exceedance levels and return intervals varied in a nonstationary analysis with time over the 54-year record, in which the location () and scale ( ) parameters were modeled as linear functions of time and the shape ( ) parameter was treated as constant following Ragno et al. (2019). To investigate how exceedance levels and return intervals varied with SOI, we performed the nonstationary analysis on the linearly detrended annual maximum sea level data with SOI as the covariate and also treated the location and scale parameters as linear functions and the shape parameter as constant. Note. The MSL "Trend" represents the mean sea level rise that has occurred over the 54-year study period. "Max range" represents the difference between the maximum and minimum sea level over the full record. "Typical range" denotes a representative range based on 2σ, where σ is the standard deviation of the sea level component over the full record.

Contributions to Sea Level Variability
To illustrate the sources of sea level variability that occur along the WA coastline, Figure 2 shows an example of the decomposition of the individual sea level contributions at Fremantle (FM). Over this 54-year period, each individual contribution to sea level variability at FM is of the same order of magnitude (10 cm). At FM, the relative mean SLR was 13 cm over the 54 years (∼2.5 mm year −1 on average), with other sites experiencing a similar rise of 8-13 cm over this period (Table 1). We note that in Table S1, we also place these relative rates into absolute rates of sea level change through trends calculated from the satellite altimetry record during a shorter period when data is available , in which we generally obtain comparable rates among all sites (within uncertainty estimates). Interannual contributions to sea level variations are large (maximum range up to ∼40 cm), with seasonal variations also significant (∼20 cm range) (Figure 2b). The monthly maximum tidal elevation varies between ∼20 and 40 cm (Figure 2e), with an interannual modulation of the tidal range (∼10 cm) apparent that is associated with 18.6-year lunar nodal cycle (Eliot, 2010; LOWE ET AL. 10.1029/2020EF001620 6 of 16 The tidal contribution (η T ) varies along the coast but is generally small along the western and southern coasts (i.e., at sites ES-CN), with a typical range (defined as  2 T , where  T is the standard deviation of the tidal contribution) being  50 cm (Table 1). A minimum tidal range occurs in the southwest (FM and BU sites) where the typical range is only 32 cm. The tidal contribution becomes considerably larger in the northwest at PH, with a typical range of ∼250 cm.
The storm surge contribution (η SS ) generally follows an opposite pattern to the tide, with the typical range of surge variability largest along the western and southern coasts (typical range 20-23 cm), and decreasing in the northwest (CN and PH) ( Table 1). At PH in the tropics, the typical surge range is only 12 cm; however, the site also occasionally experiences the largest surge contributions during episodic tropical cyclone events. Variance-preserving spectra of the storm surge contribution show that the variance is most concentrated at time scales of 1-2 weeks (Figure 3b).
The variance in atmospheric pressure shows a general decreasing trend from south to north, with the variance peaking at similar 1-2-week time scales, which falls within the "synoptic weather band" that is known to be dominant at 1-2 weeks along the WA coast (e.g., Lowe et al., 2012;Smith et al., 1991). The peaks in atmospheric pressure variance occur at slightly shorter period, which is consistent with the order hours-to-day lag between atmospheric pressure and surge that is often observed at other coastlines (Ponte, 1994;Woodworth et al., 2019). The static sea level response (η IB , inverse barometer effect) predicted from local atmospheric pressure anomalies (see Section 2.2) is strongly correlated with the storm surge contribution (η SS ) along the southern and southwestern coasts from ES to GN (   0.7 0.8

IB R
) but decreases to  0.3 IB R in the northwest (PH) where the surge variance is also smallest (Figure 3b). However, a linear regression of η IB and η SS ( Figure S2) indicates that the simple sea level response estimate of η IB underpredicts the absolute magnitudes of η SS by 50%-60% in LOWE ET AL.
10.1029/2020EF001620 7 of 16  the south and southwest (ES to FM) and by as much as 80% in the northwest (∼80%). Therefore, while the consistency in time scales between the atmospheric variability and η SS indicates that local atmospheric forcing is a major driver of the storm surge contribution, other more complex sea level responses (e.g., dynamic sea level responses to the atmosphere and wind setup) are also likely major contributors to η SS . Finally, variability in the surge contribution each year (annual standard deviation) is generally poorly correlated with SOI (Table 3), with correlations R surge-SOI  0.2 at sites in the south and southwest (ES to FM) and reaching only 0.35 in the north (PH). For the southwestern region of Australia in particular, interannual variability in storm activity is known to be shaped by other climate indices, especially the Southern Annular Mode (Cuttler et al., 2020;Raut et al., 2014).
The seasonal sea level cycle (ensemble-averaged monthly mean sea levels) reaches maximum values ∼3 months earlier along the north-western coast (PH) during the austral autumn (February-March), compared to along the southwestern and southern coast (from FM to ES) where maximum values occur during the austral winter (May-June) (Figure 4a). For sites from GN and further south, the frequency of extreme events per month displays a similar pattern to the monthly mean water levels (Figure 4b  peak coincides with the peak in the monthly mean water level at PH (similar to the other sites). The anomalous peak in September-November is due to the dominant role of tides at PH, with a cycle of perigean spring tides (i.e., "King Tides") occurring during this time of year that acts to slightly elevate the spring tidal range by 10%-20% relative to annual mean values (not shown).
The regional patterns in the seasonally averaged sea level anomalies obtained via satellite altimetry (Figure 5) indicate that the annual sea level cycle is amplified adjacent to the coast (within a few 100 km from the coast) and shows a generally consistent response along the entire WA coast (for individual monthly averaged maps, refer to Figure S3). There is some phase lag evident in the minimum/maximum values between the northwest and southwest sections of the coast (with an abrupt shift occurring near the North West Cape near 22° S); which is particularly pronounced during the austral summer months (December-February) (Figure 5a). These patterns are consistent with the monthly averaged sea level records (Figure 4a   CN and PH (either near or to the north of the North West Cape) the annual sea level cycle leads the sites in the southwest by ∼3 months.
Long-term variations in sea level (annual-averaged, detrended values) are substantial and approximately in phase along the entire coast (Figure 6b). These variations are strongly correlated with SOI ( Figure 6a) with R SOI ∼ 0.8 (Table 1), with elevated sea levels during La Niña periods (i.e., SOI > 1) and lower water levels during El Niño periods (i.e., SOI < −1). The interannual variations in water level are similarly strongly correlated with MEI (R MEI ∼ −0.8, Table 1). Given the consistency in relationship between SOI and MEI, all results that follow will focus on SOI. Maps of regional sea level averaged during La Niña (SOI > 1) and El Niño (SOI < −1) periods reveal a large-scale response (scales of order hundreds to thousands of km) within the south-eastern Indian Ocean off WA (Figure 7).

Drivers and Trends of ESL Events
Using Fremantle (FM) again as an illustrative example, Figure 8a reveals a clustering of years with a high frequency of extreme events, followed by a clustering of years when only a minimal number (or no) extreme events occur (individual results showing the trends at all sites are included in Figures S4-S9). There is also an apparent increasing trend in the annual frequency of extreme events over the 54-year record that becomes apparent in the 5-year moving average results. There were many years prior to approximately 2002 when no extreme events occurred (12 years out of this initial 36-year portion of the record) ( Figure 8). However, after 2002, extreme events have been occurring each year (2-5 events per year based on the 5-year moving average), including reaching a maximum of 10 events in 2011 that coincide with a large La Niña.
For every year when at least one extreme event occurred, it is possible to compute how individual sea level components contributed to an extreme event (expressed as the mean percent contribution to the extreme events for each year). At FM, both tide and surge make the greatest contribution to each extreme event (typically averaging 30%-50% and 20%-30% for tide and surge, respectively, over the record). The fact that tides make an important contribution at FM despite its microtidal range simply means that the extremes (defined based on hourly water levels) tended to occur within a day during higher stages of the tidal cycle. Over the 54-year period, the mean sea level trend makes an increasingly important contribution to extreme LOWE ET AL.
10.1029/2020EF001620 10 of 16 events, contributing ∼15% by the end of the record. The seasonal sea level cycle contributes on average 10%-20% to the extreme events, and while periodic, indicates that the extremes tended to occur during elevated portions of the cycle, consistent with the phasing agreement between Figures 4a and 4b. The interannual contribution, while generally small (on average contributing up to 20% over the record), plays an important role in triggering the clustering of extreme events. Years with a relatively high (low) number of extreme events tend to occur when there are positive (negative) interannual water level contributions. An example of a strong positive interaction occurred during the 2011-2013 period, when a strong La Niña occurred that elevated mean sea levels along the WA coast by ∼0.2 m; when combined with the mean SLR that occurred since 1966, this generated a high frequency of extreme events during 2011-2013. Conversely, during the strong El Niño during 2015-2016 a negative interaction occurred, where the reduced interannual mean sea level contributed to reducing the frequency of extreme events, thus playing a role in countering the influence of mean SLR.
We note that interannual variations in the frequency and magnitude of ESLs within a given year may also be influenced by interannual modulations of the tidal range over the record due to the 18.6-year lunar nodal cycle (Figure 2e). To quantify the influence of these tidal modulations relative to interannual mean sea level variations, we assess how both the annual maximum sea level and the annual number of ESL events correlate with both the interannual contribution to mean sea level and the phase of the 18.6-year tidal cycle (the latter based on computing the annual standard deviation of the tidal contribution to define a tidal range envelope). The annual maximum sea level is moderately correlated with the interannual sea level contribution (R interannual ∼ 0.3-0.5 across the sites, Table 2), but not significantly correlated with the tidal range envelope ( env tide R generally <0.2, p > 0.05, Table 2). The annual number of ESLs is even more strongly correlated with the interannual sea level contribution (R interannual generally 0.5-0.6, Table 2), but again not significantly correlated (p > 0.05) with the tidal range envelope ( env tide R generally <0.2, Table 2). Some regional differences in the contributions to extreme events are also apparent along the WA coast, which are reflected in the mean contributions to the annual maximum event (Figure 9a) as well as the overall (single) maximum event observed over the 54-year period (Figure 9b). In the far north at PH where the mean tidal range is large (∼3 m), tides make the greatest percent contribution to an annual maximum sea level event (Figure 9a). In the southwest from GN to BU, the contributions of both the seasonal and interannual water level variations are relatively large (∼20% contribution). At these locations, tides make a lesser contribution and atmospheric surges a greater contribution. The patterns of the contributions to the total (overall) maximum event are generally similar (Figure 9b).

Changes to the Probability of Extreme Events
Using data from FM, Figure 10a shows how the return level curves (computed using the nonstationary extreme value analysis described in Section 2.3) have changed at regular intervals during the 54-year period, revealing a continuous increase in the return level magnitudes at a given return period over time. This includes a projected return level curve at 2048 assuming trends in the ESL statistics continue at the same rate (Figure 10a). These results can also be expressed as effective return level curves that show how events of a given return period evolve over the record (Figure 10b). For example, based on these lines of constant return period, the 50-year event predicted in 1968 near the start of the record (1.12 m AHD) is approximately equivalent to a 10-year event in 2018.
Given the strong relationship between interannual variability in sea level and ENSO state, it is also possible to investigate how return level curves are predicted to differ between La Niña and El Niño years, that is, by replacing time with SOI as the covariate (Figure 10c). There are large differences in the return level curves between periods classified as having La Niña (SOI > 1) and El Niño (SOI < −1) conditions. The 50-year LOWE ET AL. Note. Correlations are expressed for both the annual maximum sea level and the annual number of extreme events, which are related to the annual-averaged interannual sea level contribution and the annualaveraged state of the tidal envelope, R interannual and env tide R , respectively.
Italicized correlations are significant to 95%; bold correlations to 99%.  (Figure 10d). Interestingly, the vertical offset in the return level curves between El Niño (SOI < −1) and La Niña (SOI>1) conditions (difference in return level magnitudes of  Δ 0.2 ENSO RL m) is larger than the differences to the curves of 50 yr ΔRL ∼ 0.1 m that have occurred over the past 54 years due to mean SLR (Figure 10c); and in fact more comparable to the difference predicted between the start of the record and at 2048 (

Discussion
By decomposing the individual contributions to sea level variability within tide gauge and satellite altimetry records, our results have revealed how different atmospheric and oceanic processes (over a wide spectrum of time scales) contribute to driving ESL events along the WA coastline. The relative importance of these processes varies spatially along this extensive stretch of coastline, in part due to large differences in tidal range between the north and south, as well as moderate differences in storm surge variability (Table 1). The appreciable magnitude of the low-frequency (seasonal and interannual) sources of sea level variability is particularly noteworthy and more uniformly distributed along the entire coastline, which leads to sustained periods of anomalous sea level (up to tens of centimeters) that can have a profound influence on the occurrence of ESLs. These low-frequency sea level contributions are especially important along the microtidal southwestern region where they are of comparable magnitude to the tide.
The seasonal and interannual fluctuations in sea level are closely linked to the dynamical state of the Leeuwin Current, thus revealing an interesting connection between the probability of ESL events with the largescale dynamics of an ocean boundary current system in this region. This contribution of sea level variability arises from remote generation sources of the Leeuwin Current in northern Australia that are primarily driven by the seasonal reversal of monsoon winds that initially create a large pulse of elevated sea level within the Gulf of Carpentaria that then propagates counter-clockwise around Australia toward Tasmania; as well as a secondary contribution from the large seasonal cycle of surface heat fluxes on the northwest shelf of Australia (Ridgway & Godfrey, 2015). The amplitude of the seasonal sea level cycle along WA (0.10-0.15 m) is thus relatively large when placed in a global context; however, there are also other regions that experience seasonal cycles of similar magnitude; for example, off India (Dhage & Strub, 2016) and in areas within the South China Sea (Amiruddin et al., 2015). However, for the southwestern region of Australia, where the equivalent mean tidal amplitude (i.e., half of the range) is only ∼0.2 m, these seasonal water level fluctuations make a significant contribution to the occurrence and timing of ESL events each year. LOWE ET AL.  Figure S2). The correlation (R surge-SOI ) between the annual surge variability (standard deviation each year) and SOI. Italicized correlations are significant to 95%; bold correlations to 99%. Interannual sources of sea level variability were likewise found to play a major role in driving the clustering of ESLs within particular years, which can be associated with two main drivers: (1) extended periods of anomalous mean sea level driven by the ENSO cycle and (2) the phase of the 18.6-year lunar nodal tidal cycle that regulate the maximum tidal ranges that occur in a given year. Our results indicate that the former (mean sea level anomalies) played a much more important role in driving interannual variability in ESLs (Table 2) over this 54-year study period. ENSO is a dominant global driver of interannual climate variability, which drives particularly large sea level anomalies (  20 30 cm) across the tropical Pacific that increases coastal flooding risk (Muis et al., 2018). A portion of the energy associated with these Pacific sea level anomalies becomes transmitted into the Indian Ocean through the Indonesian seas, where coastal Kelvin waves transmit these anomalies along a waveguide along the WA coast (Cai et al., 2005), which in turn modulates the strength of the Leeuwin Current (Feng et al., 2003). Our results indicate that these ENSO-driven sea level anomalies have a particularly profound influence on coastal flooding risk along the microtidal regions of the southwest coast of WA; for example, increasing the 100-year ESL event by ∼20% between El Niño and La Niña conditions (Figure 10d).

Table 3 Maximum Correlation (R IB ) and Corresponding Lag Time (Lag IB ) Between the Storm Surge Contribution (η SS ) to Sea Level Variability and the Sea
The interannual modulation of tidal range driven by the 18.6-year lunar nodal tidal cycle is known to be relatively large along the southwestern WA coast (Eliot, 2010;Haigh et al., 2011), causing tidal ranges to vary by 20%-30% (Figure 2e). This cycle last peaked along this coast in ∼2006 and subsequently reached a minimum near the end of the study period in ∼2015 (also coincident with El Niño conditions). However, the results suggest that the phase of the 18.6-year tidal cycle was a poor predictor of the likelihood of ESLs occurring within a given year, with interannual variations being much more strongly associated with EN-SO-driven mean sea level variability. Nevertheless, we do acknowledge that the length of the 54-year record may also mask a secondary role of decadal variability in tides, given the clearly large influence of ENSO LOWE ET AL.
10.1029/2020EF001620 13 of 16 (including major La Niña and El Niño during the latter decade of the record). Looking into the future, the next peak in lunar nodal tidal cycle will occur in ∼2025, at which time any elevation of mean sea level due to La Niña conditions may put the WA coastline at elevated coastal flooding risk.
Over the 54-year study period, mean sea level rose by a total of 8-13 cm along the WA coastline (Table 1), or at an average rate of ∼2-3 mm year −1 , which is comparable to the global mean rate of SLR over the same period (Church et al., 2013). This mean sea level trend made an increasingly important contribution to ESL events, especially along microtidal regions of the WA coast (i.e., from Fremantle to Esperance) (Figures 8b  and 9b). By the end of the study period (end of 2019), ∼20% of the contribution to ESL events at Fremantle can be attributed to the mean SLR over the period. Based on the nonstationary return level analysis for Fremantle (Figure 10b), the mean sea level trend has increased the effective return level for the 100-year event by ∼10%.
While the present study focuses on a hindcast analysis of the mechanisms responsible for driving ESLs (including temporal and regional trends), the results can provide insight into the trajectory of ESLs into the future. Given that the various individual contributions to ESLs can be forecast with varying degrees of skill over different time scales, our work has important implications and provides opportunities for predicting coastal flooding risk for this region into the future. Given the importance of low-frequency (seasonal and longer) contributions to ESL events along this coast, many of these contributions can be accurately forecast far in advance. At decadal time scales, the role of long-term variations in tidal range driven by the 18.6-year lunar nodal tidal cycle can be accurately forecast indefinitely into the future based on tidal harmonics alone. Similarly, mean SLR can be forecast relatively accurately in the near-term (e.g., of order decades) (Church et al., 2013). The important role of seasonal to interannual sea level contributions to ESLs also provides some unique opportunities for making robust forecasts of periods of elevated coastal flooding risk well in advance. Miles et al. (2014) and McIntosh et al. (2015) evaluated the ability of a global sea level forecast model (spanning Pacific and Indian Oceans) to predict sea level anomalies at various lead times (assessing periods up to 8 months in advance in those studies). These studies specifically identified the WA coastline as one of a few regions worldwide having a high forecast skill (correlation skill >0.6) over long lead times of 6-8 months (see Figure 3 in Miles et al. [2014] and Figure 3 in McIntosh et al. [2015]). Of all mechanisms considered, only the atmospheric surge contribution cannot be forecast with long lead times, given its dependency on resolving individual storms in the weather band that can generally only be robustly predicted up to order a week in advance. Nevertheless, storm surges alone are not primarily responsible to ESL events along this coast, contributing on average ∼25% across all sites to an annual maximum event (ranging from a 10%-40% contribution depending on site). On this basis, the development of future coastal flooding risk models (including practical Early Warning Systems) for this coast could take advantage of the fact that most of the processes responsible for ESLs can be accurately predicted months in advance, which could then combine probabilistic approaches to incorporate uncertainty around the storm surge contribution.

Conclusions
The analysis of long-term sea level observations along the extensive WA coastline has revealed how different sources of sea level variability interact to regulate the frequency and trends in ESL events, including how regional gradients in the dominant met-ocean drivers of such events (e.g., ranging from macrotidal in the north to microtidal in the south) explain patterns along the coast. Long-term sources of water level variability (seasonal to interannual) were identified as making particularly large contributions to ESLs, which are both large (relative to typical global averages) and can make a significant contribution to the total sea level variability, especially in the southwestern region of Australia where tidal ranges are small (microtidal). Within this southwestern region, all of the major individual contributions to ESLs (tide, surge, seasonal, and interannual) tend to vary by a similar magnitude (of order 10 cm, Table 1), making it important to understand how all of these factors contribute to historical trends in ESLs and how the frequency of ESLs will likely change into the future. With SLR over this 54-year study period (∼10 cm) now of similar magnitude to other sources of sea level variability, SLR was also identified as making an increasingly significant contribution to modern trends in ESLs. Overall, due to the comparatively large magnitude of long-term sea level contributions (seasonal, interannual, and longer), the WA coast can provide a useful case study of how sustained periods of elevated sea level will impact coastlines worldwide more generally.
While the focus of this study has been on examining the historical drivers of ESLs, the results provide a foundation for the development of new tools to predict how the frequency of ESLs will evolve in the future. Given the ability to forecast these individual contributions to sea level in advance range from effectively limitless (i.e., tides and the annual cycle) to order days to weeks (i.e., surges that depend on resolving individual storms), these uncertainties in individual contributions will influence uncertainties in predictions of ESLs in very different ways. Recent developments in tools to predict future coastal flooding by ESLs (e.g., Anderson et al., 2019;Vitousek et al., 2017;Vousdoukas et al., 2018) describe promising approaches to integrate deterministic predictions with probabilistic forecasts of individual contributions (each with varying levels of forecast skill). As the present study focused only on drivers of ESLs based on offshore total sea level (still water level) variability, there are also opportunities to consider all contributions to total water level variability at the coastline by incorporating predictions of wave runup (e.g., Melet et al., 2018;Serafin et al., 2017). Given the complex nature of the WA, with numerous offshore reef systems (ranging from coral to rocky) that provide significant sheltering from offshore waves, the development of tools that include wave runup will be more complicated compared to open sandy coastlines where such analysis has previously focused.
Finally, the knowledge developed to understand coastal flooding risk along WA can support future studies of the drivers of erosion along this coast, given the strong connections between ESLs and coastal erosion. Recent studies of coastal erosion along WA have highlighted the important role that seasonal and interannual water levels can play in driving shoreline variability (Segura et al., 2018). This is especially the case for the large portion of the WA coastline that receives some degree of wave attenuation from reefs and other coastal topography, which in turn reduces the influence of wave runup on beach erosion cycles (from individual storm to seasonal variations). Overall, the WA coastline can act as a valuable test bed for assessing how coastlines generally respond to sustained sea level changes, due to the major influence of long-term sources of sea level variability.