Long-term drivers of shoreline change over two centuries on a headland-embayment beach

Shoreline evolution over the last two centuries was analysed for Dundrum Bay, Co. Down, Northern Ireland, using historical and recent shoreline datasets from 1833 to 2020. The area of interest comprises two sandy beaches and vegetated coastal dune fields, Newcastle – Murlough and Ballykinler, separated by an inlet channel which connects the inner with the outer bay. Twenty-four temporal shorelines were extracted, and a quantitative assessment of their positional uncertainty was performed. This was combined with analysis of foredune volume variations by applying the Structure-from-Motion-Multi-View-Stereo technique to 1963 aerial photography and comparing it with a 2014 Light Detection And Ranging (LiDAR) dataset to better inform on links between the sediment dynamics and the observed shoreline changes. Storm events were identified using recorded extreme water levels (EWLs) (1901 – 2020) and hindcasted wave data (1948 – 2020). In the first 87 years, the shoreline was largely stable, and change was focused at the inlet area. In the 20th century, localised retreat characterised the western (Newcastle – Murlough) sector, whereas the eastern sector (Ballykinler) experienced general shoreline advance. The rate and extent of shoreline retreat in the western sector increased post-1997 in concert with accelerated accretion at Ballykinler. The strongest erosional episodes were recorded in 1920 – 1951, 1997 – 2005, and 2012 – 2014. Although no direct link was established between single storms and shoreline retreat rates, the three major retreat periods coincided with several consecutive EWLs or EWLs with a return period greater than 100 years. These were generated by storm directions ranging from south to south-west. The long-term pattern of shoreline change points to a complex coastal system that is still evolving.


| INTRODUCTION
In coastal geomorphology, the longest field-measured topographic datasets are the JARKUS dataset in Holland, which consist of annual beach profiles taken every 250 m along the entire Dutch coast since 1965(de Vries et al., 2012;;van IJzendoorn et al., 2021) and the Narrabeen-Collaroy beach in Australia, where almost 50 years of monthly cross-shore profiles have been measured since the late 1970s (Turner et al., 2016).These datasets are rare exceptions worldwide, demonstrating just how difficult it is to maintain continuous and consistent in situ observations as different data acquisition techniques and instruments evolve (Splinter et al., 2018).
Other types of long-term datasets incorporating a measurement of shoreline position have emerged and are mainly derived from satellite observations beginning from the late 1970s (Mentaschi et al., 2018).Extracting shoreline position or geomorphic information from historical maps and vertical aerial photos, however, remains the only way to produce datasets extending over several decades (Fellowes et al., 2021;Talavera et al., 2021;Thomas et al., 2011;Tõnisson et al., 2012) and in some cases for more than one century (Latapy et al., 2019(Latapy et al., , 2020;;Pollard et al., 2020).
Long-term analysis of coastal change allows identification of periodic coastal patterns and hotspots of erosion which can occur over several decades (e.g., Hein et al., 2019) and thus are crucial for longterm management of coastal environments.In reality, adaptation strategies of coastal areas are often conceived without a long-term perspective, despite the long-term impact of climate change on these areas (Horstman et al., 2009a).Decision-makers can be conflicted between the warnings of the scientific community and the desire to give effective and immediate responses to the concerns of local coastal communities; thus, the usefulness of long-term coastal management strategies is still debated due to the inherent physical and sociological uncertainties involved (Horstman et al., 2009b).
Recently, attempts have been made to model complex shapes of shorelines including future predictions of shoreline change up to century timescales (Roelvink et al., 2020).These models however, are still only validated using field-surveyed data and observations at just a limited number of test sites and over restricted timescales of short-to medium-term (few years) periods (Scardino et al., 2020).Shoreline variations at decadal to century timescales have been attributed to various factors including changes in shoreface morphology, with consequent modifications of the hydrodynamic forcing (Backstrom et al., 2009;Hamon-Kerivel et al., 2020, 2023); changes in the longshore sediment transport gradient; variations of sediment sources and sinks; tectonic activity; human interventions; and sea level rise (Cooper et al., 2020;Le Cozannet et al., 2019;Montaño et al., 2020).
Currently, the only model of coastal change that directly uses historical shorelines to better refine future projections is CoSMoS-COAST (Vitousek et al., 2017).
Wave conditions and water level at the shoreline are the two relevant components of coastal storms, as well as the type of coast plays a major role in how a shoreline responds (Cooper et al., 2004;Guisado-Pintado & Jackson, 2018;Harley, 2017).Local wave energy as well as storm frequency influences actual shoreline response (Angnuureng et al., 2017).Other storm characteristics such as the duration and storm track can diversely impact different coastal locations (Guisado-Pintado & Jackson, 2019;Rulent et al., 2021).Furthermore, depending on the characteristics of each coastal setting, a major influence on shoreline change is the synchronicity of local forcing factors and the previous beach conditions (Crapoulet et al., 2017;Guisado-Pintado & Jackson, 2019;Keijsers et al., 2014;Splinter et al., 2014) or when the effects of a storm (or its most energetic part) coincide with high tides at the shoreline (Pye & Blott, 2008).Wave direction and energy are also influenced by geological constraints such rocky headlands which affect wave shoaling and shoreline response to storms in headland embayed beaches (Daly et al., 2014;Gallop et al., 2020).Over a long period of time, an important contribution to shoreline change is possible through by the occurrence of tsunami events where these events can be a significant factor in modifying the coastline (Scicchitano et al., 2022).
Here, we document long-term shoreline changes and identify the main drivers of retreat/advancement phases in a low energy and sheltered beach that is also influenced by anthropogenic impact over the decades.We examine long-term shoreline variation from multiple

| STUDY AREA
The study site is located at Dundrum Bay, in the southeast of Northern Ireland along the Irish Sea coastline (Figure 1a).The bay is delineated by the Mourne Mountains in the southwest and by St. John's point in the east (Figure 1b).Dundrum Bay is one of the largest embayed beach systems in Northern Ireland with a large and vegetated dune field dating back to the mid-Holocene (Orford et al., 2003;Orford & Murdy, 2002).The wide sandy beach is characterised by Multiple Inter-Tidal Bars (Biausque et al., 2020).An extensive ebb tidal delta divides the southwest sector (Newcastle and Murlough) from the eastern sector (Ballykinler) (Figure 1b,c).In the Newcastle-Murlough side, a steep gravel ridge extends from the dune toe to the high tide mark (Cooper & Navas, 2004).The tidal environment is high mesotidal with a mean spring tidal range of 3.89 m (Jackson et al., 2022).Wave climate is dominated by short-period wind waves (Cooper & Navas, 2004) with a mean significant wave height of 1 m (Jackson et al., 2022;Orford, 1989).Modal winds are from south and southwest, whereas storm winds are common from the southeast (Orford, 1989;Navas et al., 2001; Figure 1d).Carter (1988) defined Dundrum Bay as a dissipative beach with a surf-scaling parameter between 20 and 200.Newcastle has been an established seaside resort since 1837 (Hanna, 2002), and the first concerns about beach erosion emerged at the beginning of the 20th century (Andrews, 1901).Attempts to stop shoreline recession were made at different stages between 1910 and 2007 with the construction of hard defences like rock armouring, seawalls, and a promenade between Newcastle and the Royal County Down golf course (Hanna, 2002).Despite these efforts, erosion has continued through the end of the century alongside increasing recreational pressure and popularity of the site (Whatmough & Carter, 1982).
In the early 19th century, vessels of 50 tons could navigate through the inlet channel (Pye & Blott, 2017).Later in the century, sedimentation led to infilling of the channel which was then dredged periodically until the closure of the port of Dundrum in 1984 (Pye & Blott, 2017).The area was declared an Area of Special Scientific Interest in 1995 under the name "Murlough ASSI" (DAERA, 1995), and the entire bay is also designated a Special Area of Conservation (DAERA, 2007).The dune system of Murlough Special Area of Conservation has been managed by the National Trust since the 1960s (Whatmough & Carter, 1982). in all data sources and was reasonably continuous.It has been recognised as the most appropriate shoreline proxy when a good compromise between responsiveness to beach dynamics and identification from remotely sensed data is required (Pollard et al., 2019).
The main shortcoming of using the seaward edge of dune vegetation is that it may not keep pace with short-term accretion (Boak & Turner, 2005), but considering the long-term perspective of this work, it was considered an appropriate indicator.Large dune blowouts, despite being partially flooded at high tide after storms, were trimmed at their neck (entrance) and excluded from the shoreline analysis because the generation and evolution of these geomorphic features are mainly driven by wind erosion and aeolian processes.

| Shoreline change analysis
Shoreline variation was analysed using the ArcMap ® tool Digital Shoreline Analysis System (DSAS) 5.0 (Himmelstoss et al., 2018).A total of 325 transects was used at 25 m intervals.The DSAS statistics were computed using a confidence interval of 99.7% both for the entire study site and for the two subsites of Newcastle and Ballykinler.Newcastle can be divided in three zones from SW to NE: (i) golf course; (ii) Murlough; and (iii) the sand spit close to the inlet channel (Figure 1c).Ballykinler, on the other side of the inlet channel, can be divided in other three zones: (i) inlet zone 1, (ii) inlet zone 2, and (iii) the rest of the beach (Figure 1c).Statistics selected for the shoreline analyses were as follows: (i) the Shoreline Change Envelope which represents the greatest maximum difference in shoreline position on a given transect.The value for Shoreline Change Envelope is consequently always positive (Himmelstoss et al., 2018).(ii) The Net Shoreline Movement (NSM) is the distance between the oldest and the youngest shorelines for each transect (Himmelstoss et al., 2018).
Values can be positive (advancement) or negative (retreat).(iii) The End Point Rate (EPR) is calculated by dividing the distance of shoreline movement by the time elapsed between the oldest and the youngest shoreline (Himmelstoss et al., 2018).Values can be positive (metres of accretion per year) or negative (metres of retreat per year).
EPR values have an uncertainty value (EPR unc ) derived from the total error of the two shorelines considered.The total shoreline error for each year was given by different uncertainties associated with each data source from which the shoreline was digitised (Buchanan et al., 2020;Cenci et al., 2013;Crowell et al., 1991;Fabbri et al., 2021;Hapke et al., 2011;Moore, 2000;Morton and Miller, 2005;Morton et al., 2004;Oyedotun, 2014;Romine et al., 2009;Ruggiero et al., 2013;Shalowitz, 1964;Virdis et al., 2012;Tables S1 and S2).All the above statistics were also run between every single time interval   Grottoli et al., 2021).This test was performed to select the best DSM in terms of uncertainty and spatial coverage.The DPC used in this paper to reconstruct the 1963 DSM was produced by removing from it points with a confidence level between 0 and 2 (confidence level calculated by the software) and using a "moderate" filter for the depth maps.The DPC quality was selected at an "ultra-high" level in the software to take advantage of the entire pixel number of the images.For more details on how the 1963 DSM was reconstructed, see also Grottoli et al. (2021) A limit of detection (LoD) of ±0.984 m was applied on the digital elevation model of difference.The LoD was calculated as the root of the sum in quadrature of the uncertainties associated with each individual DSM (Lane et al., 2003) and setting a minimum level of detection with a confidence limit of 68%.The standard deviation of the error was used as uncertainty to calculate the LoD, with a value of 0.066 and 0.983 m for the 2014 LiDAR-derived DSM and the 1963 AP-derived DSM, respectively (Table S3).Volume calculation was performed only on areas with significant changes, as well as volume uncertainty was estimated multiplying the LoD for the area that experienced significant changes (i.e., above or below LoD) (James et al., 2017(James et al., , 2019;;Milan et al., 2011).

| Identification of storms and EWL events
Different sources of wave data were collected to identify storm events during the entire period of interest.For the period 1833-1948, sources of information are represented by news articles about storms in local newspapers collected from the British Newspaper Archive and the Irish Newspaper Archive (https://www.britishnewspaperarchive.co.uk/ and https://www.irishnewsarchive.com/).Because no physical characteristics can be estimated from these sources of information, consequences of each storm event such as shipwrecks, deaths, damages to buildings and infrastructures, or floodings are reported only.For 1948-2015, hourly hindcasted wave parameters (significant wave height, Hs; peak period, Tp; and, direction, Dir) were provided by the Environmental Hydraulics Institute of University of Cantabria (Menendez et al., 2014;Reguero et al., 2012).
Three-hourly wave parameters were provided by Ifremer (France), based on WaveWatch III hindcast model, for the remaining period 2015-2020.Both hindcast models provided wave parameters extracted from the same grid point which approximately 30 km offshore of Dundrum Bay (54.0000N, 5.5000 W; Figure 1a) in a water depth of 88 m.Hindcast wave data were validated against recorded data from the M2 buoy at location (53.4836N, 5.4302 W; Figure 1a).
In the hindcast datasets, storms were identified from a statistical point of view using the peak-over-threshold method (Harley, 2017).A storm event was identified if H s exceeded the 0.95 quantile (2.35 m for both hindcast wave datasets) for a minimum of 15 consecutive hours that corresponds to 0.75 quantile of the duration with H s above 2.35 m.
The wave energy (E) of each storm was then quantified according to Bryan and Power (2020).
For the remaining period 2010-2020, the water levels recorded at the nearby tide gauge of Bangor (Figure 1a), provided by the British Oceanographic Data Centre, were analysed.2b).Accretion dominated the eastern part of the bay (Ballykinler beach), with a maximum NSM of +208 m in the inlet zone 2 (Figure 2b).The shoreline at Newcastle-Murlough retreated at an average rate of À0.13 ± 0.09 m/year (with a peak value of À0.33 m/ year), whereas average advance on the sand spit was 0.73 ± 0.09 m/ year (with a peak value of 1.29 m/year; Figure 2c).Ballykinler beach had an average advance of +0.41 ± 0.09 m/year (with a peak value of 1.11 m/year).
4.1.2| Foredune volume variation  From 1963 to 2014, the volume loss from the Murlough-Newcastle foredune was approximately 240 000 ± 57 000 m 3 , whereas at Ballykinler, the foredune gained 495 000 ± 7500 m 3 within the same period (Figure 3a).Most of the volume lost in Murlough was from the foredune SW of the sand spit (Figure 3a).
In that zone, the foredune experienced a lowering of more than À15 m in some locations.At Ballykinler, despite not being able to reconstruct the DSM from the 1963 aerial photos, most of the volume gain was concentrated within the foredune area at the rear of the ebb tidal delta (Figure 3b), with maximum elevation gains of +12.8 m (Figure 3b).Peak values of erosion and accretion correspond to locations where peak shoreline retreat and advance were also recorded (Figure 2).4.1.3 | Storm events (overall: 19484.1.3 | Storm events (overall: -2020) ) Examining the entire period of hindcast wave data from 1948 to 2020 revealed a total of 738 storms having occurred at the site.The majority of storm directions was from southerly directions (Figure 4a and Table 1).The maximum significant wave height (H s max) was between 3 and 4 m for around 55% of the cases, 24% were between 2 and 3 m, and 17.5% between 4 and 5 m (Table 1).Around 45% of storms had a cumulative wave energy between 51 227 (min) and 250 000 J/ m 2 , 40% between 250 000 and 500 000 J/m 2 , and almost 11% between 500 000 and 750 000 J/m 2 (Figure 4b).

| Shoreline variation for each time interval
During the period 1833-1920, most of the shoreline displacement occurred along the inlet area (Figure 5a).In Murlough, the shoreline displacement was associated with generation and growth of a sand spit that caused the eastward shifting of the inlet channel and erosion on its eastern side (Figure 5a).The maximum advancement of the sand spit was between +150 and +175 m (Figure 5a), with a maximum rate of advance between +1.8 and +2.0 ± 0.1 m/year (Figure 6a).At Ballykinler, the inlet area at the rear of the ebb tidal delta started to advance with peak values between +175 and +200 m (inlet zone 2; Figure 5a).The retreating section at Ballykinler had peak values between À25 and À50 m (inlet zone 1; Figure 5a) with a retreat rate between À0.3 and À0.6 ± 0.2 m/year (Figure 6a).
The rest of the beach, both at Newcastle-Murlough and Ballykinler, experienced insignificant change (i.e., within the error margin of the dataset; Figures 5a and 6a and À63 m (Figure 5b) and a retreat rate between À0.6 and À0.9 ± 0.2 m/year were identified (Figure 6b).At Ballykinler, the beach advanced by between +10 and +100 m (Figure 5b) at average rates between +0.30 ± 0.2 and +0.90 ± 0.2 m/year (Figure 6b).
During the period 1997-2020, growth of the sand spit slowed and only one transect advanced between +1.5 and +1.8 ± 0.1 m/year (Figure 6c) and a peak shoreline advancement between +25 and +50 m (Figure 5c).The retreat trend for Murlough extended to cover the entire Murlough shoreline up to the golf course (Figure 5c).The highest retreat values were recorded in Murlough with peak values between À25 and À50 m (Figure 5c) with peak rates between À1.2 and À1.32 ± 0.1 m/year (Figure 6c) which represent the highest retreat rate recorded overall.At Ballykinler, except for a small retreating section in the inlet zone 1 (Figure 6c), the rest of the shoreline experienced rapid rates of accretion (i.e., more than +2 m/year in inlet zone 2 and between +0.3 and +1.8 ± 0.2 m/year in the rest of the beach; Figure 6c).between 1900 and 1920 (Figure 7).The rest of the foredune at Murlough beach had a constant retreat trend with periods of relative major retreat: four consecutive retreat periods (1859-1900, 1900-1920, 1920-1951, and 1951-1962)  On the other side of the inlet channel, Ballykinler experienced the largest shoreline variation in the inlet zone, where two zones were distinguished: inlet zone 1 and inlet zone 2 (Figure 8).In inlet zone 1, the strongest retreat was during 1859-1900, with a peak value of À40 m (Figure 8a).This retreat coincided with strong growth of the sand spit on the other side of the inlet channel   (Figures 5a and 7a).Despite some gaps in the dataset, phases of retreat, slightly above the error margin, were recorded in 1997-

| Storms and EWL events for each time interval
A total of 17 storm events pre-1948 (i.e., pre-hindcasted data) were reported from local journals at Dundrum Bay or close coastal locations on the western side of the Irish Sea (Table 2).
Considering the cumulative wave energy during storms divided by the years elapsed in each period limited by two consecutive shorelines, the highest value occurred in 1991-1997, followed by 1974-1979 and 1984-1990 (Figure 9).Since 2012, when shoreline images have been available every 2 years or less, the highest cumulative wave energy per year was during 2012-2014 and 2019b-2020 (Figure 9).(Figure 6b,c).This was ascribed to a drift reversal prompted by changes in the nearshore bathymetry (Cooper & Navas, 2004).The accretion trend was also confirmed by foredune volume analyses between 1963 and 2014: The foredune in Ballykinler in those 51 years gained twice the volume that was lost from the Murlough foredune in the same period of time.This suggests that the sediment filling Ballykinler is not just coming from the Murlough foredune via ebb delta bypassing but is also provided by a more complex local sediment dynamics involving subcells of sediment transport that are feeding the tidal inlet area from both sides (Biausque et al., 2023).
Furthermore, sediment input from subtidal areas cannot be excluded, and the eastern part of the study site is also delimited alongshore by the Craighalea rocks outcrop that partially prevents sediment moving any further east, facilitating sand deposition locally (Figure 1c).Sediment derived from the ebb tidal delta, in the process of inlet bypassing, was probably the main source for the natural nourishment of Ballykinler through the decades.How the ebb tidal delta affects the wave energy dissipation and wave direction needs to be better understood.Changes in ebb tidal delta volume and morphology through time would also be crucial to better resolve the sediment exchange between the two sides of the inlet channel (Murlough and Ballykinler) in the last two centuries.Cooper and Navas (2004) have previously purported that sediment accumulation (43 000 000 m 3 ) in the subtidal zone over the 20th century could have altered wave-propagation patterns and therefore the potential sediment transport at the shoreline.
The accretion of the ebb-tidal delta in the 20th century is also consistent with loss of intertidal area and reduction of tidal prism in the inner bay due to land reclamation of 19th century in the inner bay T A B L E 2 Reported storm events for the period 1833-1948 (pre-hindcast data).the west (beyond the immediate inlet area) that occurred after 1997 (Figure 6c) was not accompanied by coeval and significant changes in the inlet area or at the sand spit.This suggests a multidecadal hysteresis between major changes in the inlet area and morphological adjustment of adjacent beaches.A similar pattern was also observed by Elias et al. (2022) for a mixed-energy tidal inlet in the Netherlands over a 400 years analysis, even though the inlet area they analysed, like many others, underwent substantial changes induced by extensive coastal defence works (Castelle et al., 2007;Pacheco et al., 2008;Velasquez-Montoya et al., 2020).Multidecadal hysteresis between major changes in the inlet area and morphological changes of adjacent beaches could also have been induced by the reduced tidal prism in the inner bay during the 19th and 20th centuries due to a significant land reclamation (Pye & Blott, 2017) which contributed to the shallowing of the inlet channel and allowed sediments to bypass more efficiently, leading to the onset of Ballykinler accretion.Such erosion and accretion phases, especially when inlet dynamics are involved, can only be understood by considering the entire system of sediment dynamics on a temporal scale longer than a few decades.

| Storms and shoreline retreat
The strongest and episodic phases of coastal retreat occurred in 2012-2014, 1997-2005, and 1920-1951 (Figures 7 and 8) (Figures 7 and 8).The low temporal resolution of the shoreline dataset, especially for the first decades of our period of interest (1833 to 1951) and the lack of quantifiable coeval forcing parameters for storm events (i.e., hindcast storm data were only available since 1948 and recorded water levels since 1901), precludes analysis of clear causative links between storms and retreat phases.The periods 1920-1951 (due to the lack of storm data), 1997-2005, and 2012-2014 were not the most energetic periods in term of wave energy during storms (Figure 9), but consecutive EWLs and EWLs with a return period higher than 100 years (Tables 3 and 4) coincide with episodic coastal retreat phases.In fact, two consecutive EWL events in a single year occurred in 1946, prior to the 1951 detected retreat (Figures 7a and 8a), and again in 2014, prior to the 2014 widespread retreat of the Murlough foredune (Figure 7c), two consecutive EWLs occurred with one of them having a return period of more than 100 years (second top most EWL ever recorded; Table 3).Furthermore, the locus of shoreline retreat during 2012-2014 (i.e., the 2.5 km between blowout zones 2 and 3) (Figure 7c) is in the location of maximum wave energy dissipation (Biausque et al., 2022).
The highest energy events recorded by both storm wave energy and EWL proxies were recorded in 1991 and 2002.In 1991, despite being the topmost EWL event ever recorded and significant storm wave energy was hindcast in a coeval storm (storm 419 in Table 4), very little retreat signs were evident in Murlough for the period 1991-1997 (Figure 7b), whereas larger advances were recorded in the inlet zone 2 of Ballykinler (Figure 8b 4).That event had a return period higher than 100 years (third highest EWL recorded; Table 3), and it was also associated with a hindcasted storm (storm 563 in Table 4) with the second highest wave energy delivered (7.43E+05 J/m 2 ; Table 4) and a duration of 44 h (joint top event;  et al., 2018).
Whereas for the 1920-1951 and 1997-2005 periods, it  along the western coastline of Europe (Castelle et al., 2015;Héquette et al., 2019;Masselink, Castelle, et al., 2016;Masselink, Scott, et al., 2016;Rulent et al., 2021).In particular, the 2013-2014 winter was the most energetic on the Atlantic coast of Europe since 1948 (Masselink, Castelle, et al., 2016) and since 1953 for southwest England (Masselink, Scott, et al., 2016).Major damage and flooding were also recorded on several parts of the Irish coast like Cork, Wicklow, and Dublin (O'Brien et al., 2018), but there is no information on post-storm recovery.Some locations on the Atlantic coast of Europe already recovered after 1.5 (Castelle et al., 2017), 2, or 4 years after the 2013-2014 winter storms (Dodet et al., 2019), whereas Murlough's foredune never showed recovery in the following 6 years (Figure 7c), likewise it never recovered from the other two major retreat phases of 1920-1951 and 1997-2005.The sites analysed by those authors are all openly exposed to the Atlantic Ocean and were therefore affected by a different hydrodynamics compared with the sheltered Irish Sea system of Dundrum Bay.
Looking at storm direction, EWL events were always combined with storm coming from the south to southwest sector (Table 4) despite the frequency of storm direction is similar between the southwestern and the southeastern sectors (Figure 4 and Table 1).The higher chance to have a large storm from the southwestern sector combined with an EWL is likely due to the larger fetch available in the Irish Sea when a storm is coming from south or southwest (Figure 1a).shoreline retreat pre-2005 reflects the low temporal resolution of shoreline data, but storm energy is not the sole factor that can cause retreat in the system.For example, the 2019-2020 energy was much higher compared with previous periods since 2014 (Figure 4), but no EWLs were recorded nor significant shoreline retreat was observed in that period (Figures 7c and 8c).The major role of EWLs on shoreline retreat compared with high wave energy was already shown by Esteves et al. (2012).Héquette et al. (2019), in particular, demonstrated that significant shoreline retreat that occurred on the north coast of France during the 2013-2014 winter storms was mainly due to EWLs, even if wave energy was not as high as during preceding storms when only minor shoreline retreat was observed.The lack of direct relationship between storm energy and shoreline retreat in the study site may also reflect the need for storms to occur at high tide and/or with large storm surge to impact the shoreline (Biausque et al., 2023).
In conclusion, Murlough and Ballykinler beaches showed different responses to storms: Murlough was affected by the major retreat phases of 1920-1951, 1997-2005, and 2012-2014 and never showed recovery signs after those periods; whereas Ballykinler, apart from its inlet zone, was never significantly affected during those retreat periods (Figures 7 and 8).
Despite being a comparatively low energy and sheltered beach, with a wide, ultradissipative intertidal beach (Biausque et al., 2020(Biausque et al., , 2022)), this system responds to storm events of exceptional magnitude, similar to the 2002 event or the 2013-2014 winter storms.
These have a preferential direction from the south to southwestern sector and can produce EWL events with a return period above 100 years or consecutive EWL events that cause episodic retreat of the Murlough foredune without subsequent recovery.

| Coastal management and human impact on shoreline change
Despite the golf course area not experiencing significant shoreline retreat in the last two centuries, this does not imply that this sector of the study site is safe from future storm events.As observed by Hanna  Note: The EWL events are listed in chronological order.The EWLs from the period 1901-2010 are modified from Orford and Murdy (2015).EWL values are above Belfast datum.Abbreviation: EWL, extreme water level.
centuries compared with the golf course area (Figure 10c), can still evolve following natural shoreline dynamics.
Other human disturbances took place in the study site over the decades.As noted by Pye and Blott (2017), in 1984, the port of Dundrum in the inner bay (Figure 1c) was closed and periodic dredging in the inlet channel to allow navigation was stopped.In Ballykinler, retreat was only recorded for a brief interval in 2012-2014 (Figure 8c), and its rate of shoreline advance for most of the last two centuries accelerated between 1997 and 2020 (Figure 6c).The latter may be also linked to the cessation of dredging operations in the inlet channel because effects of human-induced forcing can be visible at a decadal scale on shoreline evolution (Stive et al., 2002).
From a management point of view, it is important to maintain and not to interfere with the natural dynamics of the foredune and the adjacent beach.Despite every study site being controlled by its own geomorphic evolution, sediment circulation, and hydrodynamics, a general increasing trend in vegetation cover was noted in Ireland (Jackson & Cooper, 2011) and worldwide (Jackson et al., 2019) due to climate change and changes in land use in the last century (Gao et al., 2020).Vegetation cover can also stabilise the dune sediment and promote new sediment accumulation thereby improving its storm-buffer function (Jackson et al., 2019).Continuing to interrupt the natural sediment exchange between the foredune and the beach by extending seawalls or rock armours will bring to the complete hard defencing of the coastline without keeping the supratidal lands safe from major storms or future effects of sea level rise.or EWL events with a return period higher than 100 years, with storms of exceptional magnitude, which appear to be the main driver of phases of shoreline change.In the study site, EWL events are often combined with a storm direction from the south to southwest sector but the identification of single storms responsible for significant shoreline retreat (or advancement) was not always possible, mainly due to the low temporal resolution of the shoreline dataset.

| CONCLUSIONS
datasets at Dundrum Bay (Co.Down, Northern Ireland, UK) over the last two centuries (1833-2020) and compare shoreline variations with hindcast wave datasets available for the period 1948-2020 with recorded extreme water levels (EWLs) for the period 1901-2020.Three-dimensional data and volumetric change of the foredune are also quantified from 1963 to 2014 to better understand local sediment dynamics.

3
| DATA AND METHODS 3.1 | Shoreline analysis 3.1.1| Shoreline proxy identification Twenty-four shorelines were manually digitised in ArcMap ® 10.5.1 using a dataset of historical maps, vertical aerial photographs, orthophotos, and Differential Global Navigation Satellite System (DGNSS) surveys covering the period 1833 to 2020 (Table S1).The area of interest stretched from the Royal County Down golf course at Newcastle (west) to the Craighalea rocks in Ballykinler (east) (Figure 1c).A quantitative assessment of the positional uncertainty associated with each data source was performed (see Supporting Information S1).The seaward edge of the dune vegetation was considered as a suitable shoreline proxy, because it was identifiable F I G U R E 1 Study area location.Dundrum Bay, hindcast grid point, and wave buoy (a).Physical setting of Dundrum Bay on a background image from Google Earth Pro (b).Detailed physical setting of the study site on a background orthophoto of 13 October 2005 provided by Ordnance Survey of Northern Ireland (OSNI) (c).Typical wave climate of the Irish Sea based on recorded data from the M2 buoy of The Irish Marine Data Buoy Observation Network for the period 2001-2019 (d).[Color figure can be viewed at wileyonlinelibrary.com]

(
photos with a maximum shoreline uncertainty of ±10 m); and(iii) 1997 to 2020 (period of orthophotos with a maximum shoreline uncertainty of ±4.6 m).

3. 2 |
photos.The 1963 Digital Surface Model (DSM) was reconstructed using the commercial software Agisoft Metashape Pro ® (version 1.6.3;64 bit).The technique was applied to six aerial photos dated 1963, one of the oldest available sets of photographs with a good quality, a good (30% forward and 70% lateral) overlap (i.e., the same area needs to be visible in at least three photos), and a good spatial representation of the study site.The image alignment in the software was performed by selecting the highest option as accuracy to generate a denser sparse point cloud (a.k.a., tie points).The sparse point cloud was manually cleaned, thereby removing points located outside the Ground Control points configuration.Control points were identified in locations where changes were minimal over the five decades (cross-roads, crossing trails in the dune, parking areas, etc.), and 17 of them, with a good spatial spread across the study area, were used (FigureS1).The camera optimisation and the estimation of parameters in the bundle adjustment were automatically performed by the software because no calibration certificate nor flight information were available alongside the aerial photos.Twelve different Dense Point Clouds (DPCs) were originally generated to test different combinations of depth maps filtering and removal of points with low confidence values (seeGrottoli et al., 2021).This test was performed to . The 1963 DSM was then validated against a 2014 airborne Light Detection And Ranging (LiDAR) dataset by comparing the elevation on 512 validation points well spread across the area of interest (Figure S1) that gave a standard deviation of the error of 0.983 m and a mean error of 0.934 m.The validation points were selected on zones of no change.A digital elevation model of difference between 1963 and 2014 was produced only for the Newcastle-Murlough area, as the Ballykinler site had poor overlap between aerial photos.In the Newcastle-Murlough area, volume variations were calculated from the lowest elevation available among the 1963 and 2014 DSMs (À0.504 m above Ordnance Datum Belfast) in the area delimited between the 1963 and 2014 shorelines.In the Ballykinler area, the volume calculation was performed from the lowest elevation available on the 2014 LiDAR-derived DSM (2.243 m above Ordnance Datum Belfast) in the area limited by the 1963 and 2014 shorelines.

|
Results for the entire study period 4.1.1| Shoreline variation (1833-2020) The total shoreline displacement, with no reference to a specific period, ranged between 7 and 241 m, with both these values located in the inlet area.Recession dominated in the western part of the study site (Newcastle-Murlough beach), where maximum NSM was À63 m (Figure 2b): Negative values increased towards the inlet where 90% of transects showed an overall recession.The remaining 10% of positive values are related to the generation and growth of a sand spit in the inlet (Figure ).During the period 1920-1997, the sand spit continued to advance eastward with slightly lower peak values (between +125 and +150 m; Figure 5b) and slightly lower advancement rate (between +1.5 and +1.8 ± 0.2 m/year) and over a smaller area (Figure 6a,b).A widespread retreating trend affected Murlough with values increasing towards the inlet (Figure 5b).Here, peak retreat values between À50 F I G U R E 2 Main DSAS statistics calculated for the entire period of interest (1833-2020): (a) DSAS transects and entire shoreline envelope with main geomorphic hotspots circled in white; (b) net shoreline movement (NSM); and (c) end point rate (EPR).[Color figure can be viewed at wileyonlinelibrary.com] The only geomorphic feature that experienced advance in the Newcastle-Murlough area was the inlet sand spit.There were four significant advancement phases in this sector (Figure7a): the first phase between 1833 and 1859, when the present sand spit was not yet generated and peak values of +40 m were recorded (Figure 7a); the second phase between 1859 and 1900 with the largest advancement distance ever recorded (+135 m) related to generation of the sand spit (Figure 7a); the third phase between 1920 and 1951 with a further +125 m advancement due growth on northernmost end of the spit (Figure 7a); and the final phase between 1991 and 2016 recorded slow but constant advance of the vegetation line along the entire sand spit (Figure 7b,c).In most recent years (2016-2020), the vegetation line of the sand spit appeared quasi-stable (Figure 7c).The shoreline in front of the golf course did not experience significant changes in the dune vegetation line either in any of the survey intervals apart from an advance slightly above the shoreline error between 1833 and 1859 and a retreat slightly below the error margin F I G U R E 3 Volume change between 1963 and 2014 for the Newcastle-Murlough (a) and Ballykinler (b) sides of the study area.In (a), the area where the volume change was calculated is shown using the DEM of difference technique between 1963 and 2014.In (b), the area where the volume change was calculated is shown using the 2014 DSM enclosed between the 1963 and 2014 shorelines.DEM, Digital Elevation Model; DSM, Digital Surface Model.[Color figure can be viewed at wileyonlinelibrary.com] with peak values between À25 and À35 m in 1951 (Figure 7a,b), one in 1974-1979 (peak value À11 m), one in 1997-2005 (peak values À25 m), and the last one in 2012-2014 (peak value À35 m; Figure 7c).The retreat in 2012-2014 affected the entire extent of Murlough beach for the first time, whereas the 1920-1951 and 1997-2005 had a more limited longshore extent (Figure 7a,c).

F
I G U R E 4 Frequency distribution with the maximum significant wave height (H s max) and mean direction of the 738 storms that occurred between 1948 and 2020 (a).Frequency distribution diagram with the wave energy and mean direction of the 738 storms occurred between 1948 and 2020 (b).[Color figure can be viewed at wileyonlinelibrary.com]T A B L E 1 Frequency distribution of the maximum significant wave height (H s max) and mean directions of the 738 storms that occurred between 1948 and 2020.
2005 and 2005-2009, which were again related to the simultaneous advance of the sand spit on the other side of the inlet channel (Figures7c and 8c).The only phase of mild shoreline advance in the inlet zone 1, even though slightly above the error margin and very limited in space, was during the period 2017-2018 (Figure8c).

F
I G U R E 5 Net shoreline movement (NSM) for the selected macroperiods: (a)1833-1920,  historical maps periods; (b) 1920-1997, aerial  photo periods; and (c) 1997-2020, orthophoto   period.Invisible transects are those locations where the NSM resulted within the error margin for each macroperiod so with no significant shoreline displacement.Transect extents are clipped according to the shoreline change envelop which represents the largest cross-shore displacement among all the shorelines intersecting a given transect for that period.The background image is the 2020 orthophoto provided by the ordnance survey of Northern Ireland.[Color figure can be viewed at wileyonlinelibrary.com]In inlet zone 2, six phases of advance and four retreat phases were observed, all of them with different magnitudes (Figure8).The first and the largest advance phase was between 1859 and 1900, with a peak value of +188 m (Figure8a).The zone remained basically stable between 1900 and 1920, and then, a significant foredune retreat appeared in 1951, with a maximum retreat value of À112 m (Figure8a).The second advance phase took place between 1951 and 1962, with a peak value of +17 m (Figure8b).A mild retreat phase F I G U R E 6 End point rate (EPR) for the selected macroperiods: (a) 1833-1920, historical maps period; (b) 1920-1997, aerial photos period; and (c) 1997-2020, orthophotos period.Invisible transects are those locations where the EPR resulted within the error margin for each macroperiod so with no significant rate of shoreline change.Transect extents are clipped according to the shoreline change envelop which represents the largest cross-shore displacement among all the shorelines intersecting a given transect for that period.The background image is the 2020 orthophoto provided by the ordnance survey of Northern Ireland.[Color figure can be viewed at wileyonlinelibrary.com]F I G U R E 7 Shoreline variation within each time interval available for the Newcastle-Murlough area.(a) Shoreline variation for the period 1833-1951 characterised by the highest error margin of ±17.35 m.(b) Shoreline variation for the period 1951-2005 characterised by the highest error margin of ±10 m.(c) Shoreline variation for the period 2005-2020 characterised by the highest error margin of ±4.60 m. [Color figure can be viewed at wileyonlinelibrary.com] characterised the area until 1979, and a third strong advance phase occurred between 1991 and 1997, with a peak value of +82 m (even though the dataset is spatially incomplete for this period; Figure 8b).A fourth phase of milder but more spatially spread advance followed between 1997 and 2005, with a peak value of +25 m (Figure 8c).In 2005-2009 and 2012-2014, two small and localised retreat phases occurred (Figure 8c), and they were followed by two comparable advance phases in 2014-2016 and 2017-2018 (Figure 8c).Constant accretion for the rest of the Ballykinler beach was recorded since 1920 (+0.86 ± 0.2 m/year) except for one period of retreat (peak NSM value À18 m and EPR À 3.22 ± 0.6 m/year) in 2012-2014 (Figure 8).

F
I G U R E 8 Shoreline variation within each time interval available for the Ballykinler area.(a) Shoreline variation for the period 1833-1951 characterised by the highest error margin of 17.35 m.(b) Shoreline variation for the period 1951-2005 characterised by the highest error margin of 10 m.(c) Shoreline variation for the period 2005-2020 characterised by the highest error margin of 4.60 m. [Color figure can be viewed at wileyonlinelibrary.com]The top 34 EWLs (1901-2010) are listed in Table 3.Three EWLs occurred in 1946 and in 2014, and the four highest recorded events took place in 1991, 2014, 2002, and 2007.5 | DISCUSSION 5.1 | Possible links between net shoreline change and inlet dynamics at Dundrum Bay Most of the shoreline change from 1833 to 2020 occurred around the inlet channel.Here, during the 19th century, the generation and growth of a sand spit in the west and accretion on the eastern margin caused narrowing of the inlet channel, a situation usually associated with a decrease in tidal prism (O'Brien, 1931).At a larger scale, the patterns of erosion and accretion indicate that sand movement has been generally eastward, noted from Newcastle-Murlough to Ballykinler.During the entire study period, the wave energy during storms was largely from the south, and this storm direction was the main driver of eastward sediment displacement given the SW-NE beach configuration.Accretion in Ballykinler's foredune started after 1920 and continued thereafter and at a faster rate after 1997

(
Pye & Blott, 2017).On similar sheltered beaches and tidal environments, O'Connor et al. (2011) also found that larger changes in the inlet area drive shoreline changes in adjacent beaches.In our study site, this was clearly observed over the 19th and 20th centuries, whereas the acceleration and expansion of the erosion zone towards ).It is likely that the lower temporal resolution of the shoreline data did not allow a clearer picture of what happened in the foredune response in 1991-1997.During the period 1997-2005, only one EWL event was recorded (1-2 February 2002; Table is difficult to find clear relationships between storms and shoreline retreat, mainly due to the low temporal availability of shoreline data; there is an evident temporal coincidence of the 2012-2014 retreat phase with the 2013-2014 winter storms and the two associated EWL events (Tables 3 and 4 and Figure 7c).Several authors already highlighted the exceptional severity of the 2013-2014 winter season F I G U R E 9 Cumulative storm wave energy of a single year for each period.Period colours match those in Figures 7 and 8 for the shoreline change of each interval.[Color figure can be viewed at wileyonlinelibrary.com] Shadowing offered by the Mourne Mountains headland could be the reason why the golf course was never affected by strong retreat signs, whereas Murlough, located more at the centre of the bay, was always more exposed to storms.The lack of relationship between storms and T A B L E 3 EWL events recorded by theBelfast (1901Belfast ( -2010;; edited from Orford & Murdy, 2014)  and Bangor (2010-2020) tide gauges.

(
2002), here, the dune vegetation line was stabilised via several hard interventions of coastal defence such as wooden groynes and sleepers from 1938 to the 1960s and rock armour along the dune base in 1992 (Figure 10b), suggesting that episodic shoreline retreat was evident prior to the construction of defences.Evidence of beach erosion are visible today even in this part of the study site, such as an erosive scarp affecting the road dividing the golf course from the Slieve Donard Hotel (Figure 10a) and the almost total loss of the covering sand from the beach at Newcastle and exposure of the underlying pebbles and cobbles.Natural sediment exchange between the golf course area's seaward edge and the adjoining beach has been prevented by artificial rock armouring.This also halted any shoreline change after its construction.The Murlough and Ballykinler areas, although more impacted shoreline displacements in the last two T A B L E 4 Comparison between EWL events recorded at Belfast and Bangor harbours from 1901 to 2020 and coeval storms identified from hindcast datasets.
Shoreline change and storm forcing were analysed over the last two centuries in Dundrum Bay.Collating various historical data sources and applying new techniques to their analysis (e.g., Structure-from-Motion-Multi-View-Stereo method on historical aerial photos) has enabled the creation of a long-term shoreline change dataset with an associated comprehensive spatial error assessment.This was analysed in association with EWL and hindcast wave energy as proxies for past storm events.Most of the observed shoreline change in Newcastle and Murlough was initially concentrated around the tidal inlet, but the locus of shoreline retreat gradually expanded westwards in the 20th century.On the other side of the inlet channel, Ballykinler experienced mainly shoreline advance.This general trend persisted in the two sites over the last two centuries with no signs of changing in the near future.Foredune volume changes over a 51 year interval confirmed that sand has moved eastward from the Newcastle-Murlough area to Ballykinler.The volumetric accumulation of sand in that foredune stretch was twice that lost from the Murlough foredune.This implies that other sources of sediment are nourishing Ballykinler and more complex sediment dynamics, linked to the influence of the ebb tidal delta, need to be better understood.The onset of accretion at Ballykinler and the expansion of the eroding shoreline at Murlough follow an initial period when the inlet was reduced in size.This may have facilitated inlet bypassing, but whatever the precise reason, it suggests a multidecadal delay between changes in the inlet area and adjustment of the planform configuration of adjacent beaches.The clearest episodic phases of coastal retreat were in 1951, 2005, and 2014, following the occurrence of consecutive EWL events

F
I G U R E 1 0 Erosive scarp (a) and rock armour at the dune base (b) at the southwestern corner of the Royal County down golf course despite the insignificant shoreline change over the last two centuries in a plan view (c).NSM, net shoreline movement.[Color figure can be viewed at wileyonlinelibrary.com]