The effect of tidal range and mean sea-level changes on coastal flood hazards at Lakes Entrance, south-east Australia

2.1. Observed water levels

Australian hourly WL observations are obtained from the Australian National Collection of Homogenised Observations of Relative Sea Level up to 2019 (ANCHORS; Hague et al. 2021). We also consider the Australian subset of the Global Extreme Sea Level Analysis version 3 dataset (Haigh et al. 2022) to assess changes in tidal range at additional locations not included in ANCHORS.

For the Lakes Entrance case study, we update the ANCHORS record with 6-min frequency data that is then filtered to hourly observations using a Lanczos-cosine filter for 2020 and 2021. This approximates the process used in the generation of the hourly data from higher-frequency observations in ANCHORS. We then subtract 0.757 m from Lakes Entrance observations to express WLs with respect to the Australian Height Datum (AHD). This is a mean WL datum used for flood mapping, flood mitigation and previous inundation modelling (Grayson et al. 2004; State Emergency Service 2012).

2.2. Harmonic tidal analysis

To quantify the role of tidal processes on WLs, and how these have changed through time, we perform a series of harmonic tidal analyses. We use the TideHarmomics package (ver. 0.1-1, A. Stephenson, see https://cran.r-project.org/package=TideHarmonics) to fit sums of sines and cosines of specific frequencies (constituents) to observed WL timeseries. We use the default options for this package, which involves fitting 114 constituents to the observed WLs. These constituents are then used to estimate WLs under average weather conditions (i.e. the tides) over some past or future period, termed a tidal analysis. The highest and lowest values in the analysis are termed the highest astronomical tide (HAT) and lowest astronomical tide (LAT) respectively.

In this study we use several different tidal analyses. For Section 3, we compute predictions for the 2002–2019 period based on constituents derived from each single year of data. This is repeated for all those years with at least 70% data availability. This produces different realisations of analysed tides for the period 2002–2019. To compute changes in tidal constituents in Section 3, a different approach is taken, following established methods for this type of analysis (Woodworth 2010). This involves fitting 18.61- and 8.85-year cycles to the data as well as a trend. Robust fitting is applied that down-weights outliers.

For Section 4, we define pre- and post-change tidal analyses for a case study location (Lakes Entrance, Victoria). These analyses both provide estimates of tides for the 2009–2021 period but using two different sets of constituents. The pre-change analysis is based on 1994–2001 observations, whereas the post-change tidal analysis is based on 2009–2021 observations. A step change in tidal characteristics occurred between 2001 and 2009 (a period of missing data). Hence, this pre-change analysis represents a counterfactual for what tides may have been like during the later period if the step-change in tidal characteristics had not occurred. In Section 4.3, tidal distributions are then determined using non-parametric kernel density estimation methods in SciPy (Virtanen et al. 2020) based on these analyses.

Due to missing data, nodal corrections are used to address the fact that the observed data do not cover a complete nodal cycle. However, comparison between analyses with and without nodal corrections exhibit average absolute differences equal to the data precision of 0.01 m. Some (small) difference between tidal heights might be expected for predictions from different years, both due to fluctuations in tidal constituents caused by river flow, sea-level variability, coastal stratification (Moftakhari et al. 2013; Devlin et al. 2014; Talke et al. 2020) or errors in the gauge measurement (Zaron and Jay 2014; Talke et al. 2020). However, a significant shift in tidal properties that is larger than the nodal cycle and persists over time represents a change in tidal constituents caused, for example, by a reduction in impedance through an inlet or a decrease in frictional damping (Haigh et al. 2020; Talke and Jay 2020), rather than MSL variability.

Skew surge is used to represent the non-tidal contributions to WLs. It is computed as the difference between daily maximum WL and daily maximum tide level (Batstone et al. 2013; Hague et al. 2022). This skew surge includes the WL contributions that can be attributed to both non-tidal coastal processes, e.g. as described by Woodworth et al. (2019), and riverine processes such as catchment-scale floods. This allows seamless consideration of compound flood events, where both riverine and coastal processes contribute to WLs (Leonard et al. 2014). The skew surge is only calculated on days where there are no missing observations over a 24-h period.

4.1. Impact-based flood levels

Lakes Entrance is a popular tourist town with a population of ~5000 people located in southern Australia’s East Gippsland region, 250 km east of Melbourne. Most of the town sits on a low-lying peninsula located in the Gippsland Lakes just inside an artificially maintained channel to Bass Strait, constructed between 1870 and 1889 (Wheeler et al. 2010). The water surface area of the Gippsland Lakes is almost 400 km² and the contributing catchment area is over 20 000 km², comprising six major rivers (Tan et al. 2008). The Lakes Entrance township is very low-lying with much of the town’s key infrastructure and private property sitting less than 1.3 m above AHD (~1966–1968 MSL) (State Emergency Service 2012). Accordingly, it has experienced many floods over its history. This has led to well-defined flood thresholds being developed for hazard assessments and warnings. Although the flood thresholds have historically been used for riverine flood hazards, they also apply to floods driven by oceanic influences. This follows from them being defined at the ANCHORS tide gauge and the WLs at this gauge being a good proxy for coastally driven flood impacts in the Lakes Entrance township (Table 1).

The minor flood threshold used in forecasting and warning services sits at 0.9 m AHD, whereas the major flood threshold is defined at 1.3 m AHD (Bureau of Meteorology 2013). Nuisance-type flood impacts have been modelled to occur with WLs as low as 0.7 m AHD (State Emergency Service 2012). We established a local impact-monitoring program, deploying a local observer (if available) to photograph flood impacts when WLs were forecasted to exceed this nuisance flood threshold. Over seven events in 2021 and one event in each of 2020 and 2018, flooding was observed in situ at 16 different locations within the Lakes Entrance township (Fig. 3, Table 1). This suggests that 0.7 m AHD is an appropriate impact-based threshold for nuisance flooding at Lakes Entrance (Hague et al. 2019). Most locations where impacts were reported experienced flooding when WLs reached 0.8 m AHD, with more isolated flooding at the 0.7 m AHD level. Flooding was observed due to two main mechanisms: storm-drain backflow and direct marine flooding (Habel et al. 2020). The backflow mechanism explains why saltwater flooding was often reported hundreds of metres from the shoreline (e.g. Fig. 3 photo 5), as observed in other jurisdictions (Gold et al. 2023).

Fig. 3. 

Upper panel: location of Lakes Entrance (black marker) with boundary lower panel denoted by black square. Upper inset shows the location of Lakes Entrance within Australia. Lower panel: locations (numbered) where nuisance flood impacts were reported during monitored extreme sea-level events during 2018, 2020 and 2021 and location of Lakes Entrance water level gauge (TG). Some examples of the types of impacts that were observed are provided. Photos taken by authors on 15 November 2021, except 3 (25 July 2021) and 6 (20 July 2021). Map backgrounds from Open Street Map.

4.2. Drivers of coastal WL and flood variability

The primary driver of WL and coastal flood variability at Lakes Entrance are astronomical tides, which arise from the predictable changes in gravitational forces in the earth–moon–sun system. The largest variability in high tide heights occurs on monthly timescales due to the spring-neap cycle (Fig. 4a). The highest high tide of the month is on average 0.25 m higher than the lowest high tide of the month. This is comparable to magnitude of the largest non-tidal contributions to WLs on annual timescales (Fig. 4e), resulting from wind setup within the Lakes (Walpole et al. 2011), flood flows from its major rivers (Tan et al. 2007) and storm surges (McInnes et al. 2009). Between-month variability is also important, with average heights of the monthly highest tides in June and July on average 0.18 m greater than those in March and October (Fig. 4b).

Fig. 4. 

Tidal and non-tidal factors contributing to observed water levels at Lakes Entrance. Tidal components show (a) daily and (b) monthly highest tides over time periods appropriate to demonstrate their periodicity. These are estimated using the 2009–2021 tidal analysis based on 2009–2022 constituents. (c) The height of the annual highest tide based on the 2021 iteration of the year-by-year analysis described in Section 2.2, (d) shows the annual mean sea level with trend (orange) computed by ordinary least squares, (e) shows the monthly maximum skew surge and (f) shows the astronomical tidal range, estimated using the maximum and minimum tidal level derived over the 2002–2021 period using the iterative year-by-year analysis (i.e. as per Fig. 1). Constituents were only computed for years with at least 70% completeness. Non-tidal components are estimated using (e) annual MSL with linear trend and (f) monthly maximum skew surge. Skew surge is computed as the difference between the daily maximum still water level and maximum tidal level, where tidal level is based on the 2009–2021 analysis.

Although smaller in magnitude, the year-to-year variability in the annual highest tide (Fig. 4c) is comparable to the observed increase in MSL over the 1974–2021 period (Fig. 4d). This trend computed as 1.23 mm year^–1, based on the annual MSL in years with at least 70% data completeness. Trends in MSL at Lakes Entrance are difficult to estimate with confidence due to a data gap of several years in the early 2000s and a lack of suitable reference series to perform data homogenisation using reference series before and after this data gap (Hague et al. 2021). The annual variability in the number of flood days appears closely linked to the 4.4-year cycles, which produces up to 0.06 m variation in tidal amplitudes between years (Fig. 4c). This results from solar and lunar declination, lunar perigee and their interactions (Ray and Merrifield 2019).

Understanding how high tide heights vary across different timescales is critical to understanding coastal flood hazards. We define a ‘flood day’ as a day in which the observed maximum daily WL exceeds the 0.70 m threshold associated with flooding of at least nuisance severity. This is based on the lowest WL associated with nuisance flooding impacts (Table 1). Flood days are computed from the underlying hourly data (Section 2.1) so some very short (<1 h) duration flood events may be missed. This is consistent with the approach of Hague et al. (2022). Considering flood days allows us to understand whether the variations in the monthly and annual average number of extreme tides and extreme skew surges lead to similar changes in flood frequencies. The flood level of 0.7 m AHD corresponds to the 97.5th percentile of daily maximum levels. Accordingly, we define an ‘extreme tide day’ as a day where the maximum tide exceeds its 97.5th percentile (0.57 m) and an ‘extreme surge day’ as a day where the skew surge exceeds its 97.5th percentile (0.30 m).

Using this approach, we show that despite having larger non-tidal contributions to daily WLs than most other locations in the ANCHORS dataset (Hague et al. 2022), variability in tides is more important in modulating nuisance flood frequencies than variability in skew surges. Since 2008, 75% of flood days have occurred in May, June, July or August at Lakes Entrance (Fig. 5a). All extreme tide days also occurred during these 4 months (Fig. 5c) but only 44% of extreme surge days (Fig. 5e). This strong relationship between high tides and floods exists because higher tides mean that smaller skew surges can lead to flooding, making flooding more likely to occur at these times (Dusek et al. 2022).

Fig. 5. 

Observed flood days by (a) month and (b) year and corresponding counts of (c, d) extreme tide days and (e, f) extreme surge days.

A similar phenomenon is also observed on annual timescales (Fig. 5b). The two years with the largest number of extreme tide days (2021 and 2013) also have the greatest number of flood days (Fig. 5d). By contrast, these years had near- or below-average numbers of extreme surge days (Fig. 5f). These observations indicate that astronomical forcing plays a key role in driving variability of flood days. This astronomical forcing is independent of SLR and weather and climate variability (Williams et al. 2016). Further, this interannual variability of tidal amplitudes is greater than that of annual MSL for southern and eastern Australia (Muis et al. 2018). This is why the production of accurate long-range forecasts and projections of coastal flood hazards will require both tidal and sea-level anomaly inputs (Long et al. 2021; Thompson et al. 2021; Dusek et al. 2022).

4.3. Changes in tidal variability

Changes in high tide heights due to changes in the tidal characteristics of a location can also change the non-tidal WL contributions required to cause flooding. This occurred at Lakes Entrance as the gradual decrease in tidal range throughout the 1980s and then stepwise increase between 2001 and 2009 (Fig. 1 and 4f). Analysis of pre- and post-change tidal analyses (Section 2.2) shows that this change has been non-linear, with two aspects of this non-linearity considered here. Firstly, changes in low tide heights have been larger than changes in high tide heights (Fig. 6a, c). The height of the highest annual tide over the 2009–2021 period averaged 0.67 m AHD, compared to 0.54 m AHD over the 1994–2001 period. The lowest tides were substantially lower however, decreasing from −0.31 to −0.49 m AHD. Secondly, the heights of the highest high tides of the month (i.e. spring tides) have increased by more than the lowest high tides of the month (i.e. neap tides) (Fig. 6b, d). Using the 2009–2021 constituent-based analysis, spring high tides were 0.11 m higher than those obtained using the 1994–2001 constituent-based analysis. By contrast, neap high tides were only 0.06 m higher when considering the 2009–2021 constituent-based analysis. This asymmetry has also been observed at other locations where tidal range changes have occurred (Pareja‐Roman et al. 2023).

Fig. 6. 

Changes in the statistical distributions of tidal heights computed for 2009–2021 based on harmonic tidal analysis using constituents over the contemporaneous (2009–2021) period (blue) and pre-change (1994–2001) period (orange). The kernel density estimates for (a) hourly and (b) daily maximum tide heights are shown. A 1-month timeseries from June 2021 is used to compare the (c) hourly and (d) daily maximum tide heights to illustrate how spring and neap tides vary between the two realisations of 2009–2021 tide heights.

How these changes in tidal variability can lead to changes in coastal flood risk can be examined using the concept of freeboard (Devlin et al. 2017b; Dusek et al. 2022). Freeboard is the vertical distance between the local flood level and the typical elevation of high tides and is equal to the non-tidal contributions to WLs required for flooding to occur. Following from its definition, increases in the height of high tides, whether by increases in MSL or tidal variability, can reduce freeboard. This means that both increases in tidal range and MSL cause flooding to occur more frequently as smaller non-tidal contributions to WLs can cause flooding, or flooding can occur due to tides alone (Ray and Foster 2016). In the context of Lakes Entrance this means that spring high tides being 0.11 m higher than previously have the same effect on flood hazards as a 0.11 m increase in MSL would over the same period. This is approximately double the observed SLR over the 1974–2019 period, as estimated by multiplying out the computed linear trend (Fig. 4d).

4.4. Drivers of changes in tidal variability

Changes in tidal variability are anomalous compared to those observed in other Australian locations. This suggests that localised, rather than regional, processes are driving changes in tidal range at Lakes Entrance. In the United States, similarly anomalous changes have been attributed to dredging (Helaire et al. 2019; Ralston et al. 2019; Talke et al. 2021; De Leo et al. 2022; Pareja‐Roman et al. 2023). Dredging influences tidal range by increasing the amount of ocean level variability that is transmitted through an entrance into an estuary. Build-up of sediment in an entrance has the opposite effect, potentially reducing the tidal range. The build-up of sediment in the entrance during the 1980s and 1990s and subsequent change in dredging regime to remove this sediment between 2001 and 2009 provides a plausible, physical explanation for tidal range changes observed at Lakes Entrance. Further, several large fluvial floods that scoured the entrance in the 1970s can explain the occurrence of tidal ranges similar to present-day values in the early part of the Lakes Entrance tide gauge record (Fig. 1).

Before 2008, dredging was undertaken by side casting dredges that operated whenever conditions allowed and moved sand from the channel a short distance to one side or the other of the dredge. During this period, sand built up in the entrance following several large fluvial floods in the 1970s that scoured the entrance. Since 2009, the local port authority has introduced and maintained a new dredging regime aimed at maintaining a channel of constant depth to improve vessel safety and port access (Gippsland Ports 2013). This present dredging regime operates to a target depth of 4.5 m below the lowest astronomical tide, using a trailing suction hopper dredge on a year-round basis (Gippsland Ports 2021). This is substantially deeper than the reported depth in the late 1990s, when a modelling study investigated ‘standard’ and ‘deeper’ channels of 1.9 and 3.0 m respectively (Walker and Andrewartha 2000). The new dredge physically moves the dredged material to a spoil ground well away from the channels, a method known to be a much more effective for maintaining navigable channels.

A more definitive attribution of changes in tidal range to dredging is not possible due to a lack of historical tide gauge data. For example, there are missing data during the period when the dredging regime changes and no pre-2009 data for other locations within the eastern Gippsland Lakes. There are also no WL data from the ocean side of, or within, the entrance channel pre-2009. This inhibits our ability to directly attribute changes in tidal range to dredging as has been done in other studies internationally (e.g. Ralston et al. 2019). Hydrodynamic modelling could help independently test the findings presented here and help isolate the likely causes of localised tidal range changes at Lakes Entrance. However, the agreement between the observed changes in tidal range at Lakes Entrance and the expected hydraulic response to changes in channel depth suggest that changes to dredging regime is the most likely explanation for the changes in tidal range.

4.5. Impact of tidal and MSL variability and changes on coastal flood hazards

4.5.2. Analysis of counterfactual timeseries

Following the flood days approach introduced in Section 4.2, we can consider exceedances of the 0.70 m threshold under the various counterfactual scenarios to determine the relative importance of SLR, tidal range changes and MSL variability on the number and intensity of flood days at Lakes Entrance since 2009. We define mutually exclusive subsets of coastal flood days by which events appear in which counterfactuals (Fig. 7a). This means we can categorise each recorded flood day as having occurred due to tidal range change, sea-level rise, sea-level variability or none or some combination of these factors (Supplementary Table S3).

Fig. 7. 

Nuisance flood threshold exceedances categorised based on exceedances in counterfactuals: (a) how annual flood days were classified with accompanying overall flood days in each category over 2009–2022 and (b) break-down of flood days by category in each year over 2009–2022. In both, black shading denotes that flood day would not have occurred without one of the mean sea-level trend (MSL trend) or tidal range change (TR) having occurred. Brown shading denotes that TR change, but not the MSL trend, can explain why a flood day has occurred. Red shading denotes that the MSL trend, but not the TR change, can explain why a flood day has occurred. Orange shading denotes that the coinciding presence of both TR changes and the MSL trends are required to explain a flood day. Grey shading denotes that a flood day could not be attributed to either TR changes or the MSL trend. These are separated into events that would not have occurred without large annual MSL variability (darker grey) and those that would have occurred in the pre-2009 tide regime due to the combined effects of the 1994–2001 tides and the observed skew surge (lighter grey).

Comparing counterfactual WL timeseries to the observed WLs shows that 107 out of 115 (93%) nuisance flood days recorded since 2009 would not have occurred without changes in tidal range or increases in MSL (Fig. 7a). Of these 107 days, 84 (79%) would not have occurred in the absence of changes in tidal range, whereas 50 (47%) would not have occurred if MSL had not increased. Twenty-one events (20%) would not have occurred without both SLR and tidal range changing. This means that 48 nuisance flood days (45%) would not have occurred without either tidal range changes or SLR, regardless of which it was. Hence, the effect of tidal range changes on the number of nuisance flood days is greater than the effect of SLR over the period of study.

Tidal range change may be the primary factor driving trends in nuisance flood days, but SLR has made flooding more severe across all days. Having SLR in addition to tidal range changes has led to a substantial increase in the maximum WLs during these events. Of the 36 days where tidal range change allows flooding to occur based on the counterfactual scenarios, the average daily maximum WL on flood days is 5.5 cm higher on average in CF1 (only tidal range change removed) than in CF3 (both tidal range change and SLR trend removed). The flood duration, the number of hours when WLs exceed the 0.70 m level, is also increased by SLR. Under CF2, 139 flood hours are estimated, whereas 326 flood hours were recorded in the observed WL timeseries. Hence, there are 187 additional hours of nuisance flooding associated with SLR. This is important as flood damage has been shown to be a function of both flood height and flood duration (Thieken et al. 2005).