Relative sea‐level changes in southeastern Australia during the 19th and 20th centuries

Rates of global and regional sea‐level rise between ~1850 and 1950 were high compared to those in preceding centuries. The cause of this sea‐level acceleration remains uncertain, but it appears to be pronounced in a small set of relative sea‐level proxy records from the Southern Hemisphere. Here we generate three new proxy‐based relative sea‐level reconstructions for southeastern Australia to investigate spatial patterns and causes of historical sea‐level changes in the Tasman Sea. Palaeo sea‐level estimates were determined using salt‐marsh foraminifera as sea‐level indicators. Records are underpinned by chronologies based on accelerator mass spectrometry 14C, radiogenic lead (210Pb), stable lead isotopes and palynological analyses. Our reconstructions show that relative sea level rose by ~0.2–0.3 m over the last 200 years in southeastern Australia, and rates of sea‐level rise were especially high over the first half of the 20th century. Based on modelled estimates of the contributing components to sea‐level rise, we suggest that the episode of rapid sea‐level rise was driven by barystatic contributions, but sterodynamic contributions were dominant by the mid‐20th century. Significant spatial variability in relative sea level indicates that local to sub‐regional drivers of sea level are also prominent. Our reconstructions significantly enhance our understanding of the spatiotemporal pattern of early 20th century sea‐level rise in the region.


Introduction
Future sea-level rise as a consequence of global warming is projected to be one of the main threats to civilization, affecting hundreds of millions of people living in coastal regions across the globe (Dangendorf et al., 2019;Oppenheimer et al., 2019).Current estimates of global mean sea-level rise  indicate a trend of 1.6 ± 0.3 mm a -1 (Church & White, 2006;Dangendorf et al., 2019;Frederikse et al., 2020;Hay et al., 2015;Palmer et al., 2021).Rates of at least 2.9-6.1 mm a -1 are predicted for 2040-2060, which may increase to 2.5-6.6 mm a -1 between 2080 and 2100 under even the lowest emissions scenario (Fox-Kemper et al., 2021).Quantifying past sea-level change using geological proxy records can contextualize current and future rates of sea-level rise and ascertain the relative contributions of natural and anthropogenic processes that are the underlying causes of historical global and regional sea-level change (Jevrejeva et al., 2010;Kopp et al., 2016).This is important when considering whether current rates of sea-level change are unprecedented, and what this may then mean for future adaptation and mitigation strategies.
High-resolution geological proxy records of Holocene and Common Era relative sea-level (RSL) changes from across the globe have mostly been derived from salt-marsh sediments (e.g.Long et al., 2010Long et al., , 2014;;Saher et al., 2015;Kemp et al., 2017a;Barnett et al., 2019;Gehrels et al., 2020).They demonstrate that rates of sea-level rise have deviated considerably from the late Holocene background rate, with many showing a sea-level acceleration between ~1850 and 1950~1850 and (e.g. Gehrels et al., 2005;;Gehrels et al., 2008;Gehrels et al., 2012;Kopp et al., 2015;Gerlach et al., 2017;Kemp et al., 2017b;Barnett et al., 2019).This acceleration occurred largely before anthropogenic forcing became dominant (Slangen et al., 2016).Temporal evolution of the natural and anthropogenically forced contributions to sea level, and thus changes in the magnitudes and rates of sea-level change over the Common Era, can be assessed by undertaking a global analysis.This is done by employing a spatiotemporal empirical hierarchical model that characterizes RSL in both space and time (e.g.Kopp et al., 2016;Walker et al., 2021).At a process level, RSL is modelled as a function of several terms including: globally uniform terms, a regional linear term, regional nonlinear terms and local non-linear terms (Walker et al., 2021).Hyperparameters are used to define priors such as timescales of RSL variability (i.e. centennial, millennial).Reconstructions of RSL, including vertical and age uncertainty, can then be compared to the RSL process and the reconstructions decomposed into the various terms.When a global database of RSL proxy data is included in the model, rates of global sea-level change can be inferred (Kopp et al., 2016;Walker et al., 2021).However, these compilations are spatially biased, with few high-resolution sea-level reconstructions from the Southern Hemisphere in comparison to the Northern Hemisphere (particularly the North Atlantic; Walker et al., 2021).
The few available proxy sea-level records from Australia and New Zealand document a rise in sea level ~1 mm a -1 greater on average compared to records from the North Atlantic region over the first half of the 20 th century (Gehrels et al., 2012(Gehrels et al., , 2020;;Gehrels & Woodworth, 2013;Kopp et al., 2016;Garrett et al., 2022), which could point to a far-field (i.e.Northern Hemisphere) melt source due to the effects of gravity, rotation and deformation (GRD; e.g.Bamber & Riva, 2010;Frederikse et al., 2020).A possible explanation may be melting of landbased ice masses that retreated from their 19 th century maxima at the end of the Little Ice Age (Marzeion et al., 2014).Indeed, early 20 th century global mean sea-level rise was largely driven by the melting of glaciers and, to a lesser extent, ice sheets (e.g.Parkes & Marzeion, 2018;Frederikse et al., 2020;Malles & Marzeion, 2021).However, the acceleration is not temporally consistent across RSL records, which may indicate other regional and local-scale sea-level processes acting in combination with the barystatic component to drive sea-level changes.For example, recent findings from New Zealand (Garrett et al., 2022) suggest a dominant role of sterodynamics (i.e. the combination of steric expansion and changes in ocean circulation) in driving fast rates of sea-level rise over the early to mid-20 th century.However, it remains difficult to definitively ascertain the causes behind the acceleration in the Australian and New Zealand records due to the limited network of long-term observational data and proxy records.In the North Atlantic, spatiotemporal hierarchical models have been used to extract the local and regional non-linear signals to better understand the drivers of Common Era sea-level change (Walker et al., 2022), but this is not yet possible for the Southern Hemisphere.
In this study, we create three new proxy-based RSL reconstructions for southeastern Australia to supplement sparse tide-gauge data, which increases the number of records from the region that can be used in regional and global analyses (cf.Kopp et al., 2016;Walker et al., 2021).We contextualize current and future rates of sea-level rise in the region by assessing sea-level variability over the past two centuries, and verify the existence, magnitude and timing of periods of rapid sea-level rise, as well as investigate driving mechanisms.

Study sites
Salt marshes are useful sedimentary archives of sea-level change as they accrete vertically in fine layers in response to sea-level rise (e.g.Gehrels et al., 2005Gehrels et al., , 2006Gehrels et al., , 2020;;Kemp et al., 2011Kemp et al., , 2017a)).Microfossil and chronological analyses of such archives yield high-resolution sea-level reconstructions (Barlow et al., 2013).We sampled three salt marshes in southeastern Australia, one each in Tasmania (Lutregala), New South Wales (Wapengo) and Victoria (Tarra) (Fig. 1).

Wapengo
Wapengo Marsh is located ~30 km south of Bermagui in southern New South Wales (Fig. 1).The salt marsh is ~0.51 km 2 in extent (Creese et al., 2009) and is located at the northern end of the Wapengo Lake estuary, which features a narrow bedrock-framed mouth called Bithry Inlet.Salinity in the estuary is ~34-37 ppt (Garside et al., 2014) and the modelled mean tidal range is 0.95 m (Williams et al., 2021).The marsh is dominated by J. kraussii, G. filum and S. quinqueflora as well as a lower abundance of T. arbuscula and Distichlis distichophylla (Australian salt grass).At the seaward end Avicennia marina (grey mangrove) encroaches into the marsh; this develops into a dense mangrove forest at lower elevations.European settlement of the area around Wapengo occurred ~1830-1840 CE (Lunney & Leary, 1988); however, an intensification of land use did not occur until 1861 CE, with the introduction of grazing and non-native plant and tree species into the area, reaching a peak towards the end of the 19 th century and early 20 th century (Lunney & Leary, 1988).

Tarra
Tarra is located along the Nooramunga coast in East Gippsland, Victoria, ~4 km from Port Albert (Fig. 1).The salt marsh is ~1.7 km up the Tarra River and ~1.37 km 2 in extent.This part of the Victoria coast is characterized by extensive sand and mud flats, tidal channels and sandy barrier islands known as the Nooramunga Islands.Tidal modelling following the approach used in Williams et al. (2021) predicts a tidal range of 1.25 m.A narrow fringe of Melaleuca is present around the marsh, which is dominated by D. distichophylla, S. repens, S. quinqueflora, J. krausii and T. arbuscula.Distichlis distichophylla and S. quinqueflora are most dominant at higher marsh elevations and J. krausii, S. australias and T. arbuscula increase at lower elevations.Similarly to Wapengo, the marsh is fronted by A. marina.Following European settlement of the area in 1841 CE, the population of the area expanded rapidly during the early 1840s and 1850s.

Field methods
We conducted an initial stratigraphic inspection along coring transects at each site with a hand-held gouge corer (0.5 × 0.03 m) to ascertain the area of the marsh with the thickest salt-marsh peat that might provide the longest sea-level record (Fig. 1).We took master cores from each site using either a wide-diameter gouge (0.5 × 0.06 m) or Russian peat sampler (0.5 × 0.06 m).All cores were surveyed relative to a temporary local survey peg using a Trimble M1 (DR2") total station and linked to Australian Height Datum (AHD) 2020 using a Trimble R4 real-time kinematic GNSS.We analysed fossil foraminifera in each master core at 1-cm resolution.Species were picked, identified and counted using the methods described by Williams et al. (2021).

Transfer functions and palaeomarsh surface elevation reconstruction
For Lutregala and Wapengo, we used weighted-averaging partial least squares sea-level transfer functions previously developed by Williams et al. (2021) to generate palaeomarsh surface elevation (PMSE) estimates for fossil samples in our master cores (see 'Transfer functions' in the Results).For Tarra, we counted foraminifera in 41 surface samples and developed a new local transfer function following the methodology described by Williams et al. (2021).We quantified PMSEs for samples with total foraminiferal counts >50 individuals (following Kemp et al., 2020) using the R package 'rioja' (Juggins, 2020).To convert PMSE estimates into RSL estimates, we took the field elevation of each sample and subtracted the PMSE estimate.To assess the reliability of resulting PMSE estimates we compared the fossil assemblages with modern ones by employing the Modern Analogue Technique in rioja.We use the 'Square Chord' dissimilarity metric (Birks, 1995) as this method is optimal for count data such as those used in this study and helps to minimize the signal to noise ratio (Kemp & Telford, 2015).We define the threshold between good, fair and poor modern analogues using the dissimilarity between the samples in the modern training set.In line with recommendations from Kemp and Telford (2015), we choose the 20 th percentile as the threshold for analogy as the diversity of the modern training sets are low.Following Watcham et al. (2013), we define 'good' analogues as samples with minimum dissimilarity values below the 5 th percentile and 'fair' analogues as those with values between the 5 th and 20 th percentiles.For further details on the derivation of the transfer functions, refer to Williams et al. (2021).
Figure 1.Map of the study sites showing A, regional-scale overview; B, sub-regional overview of the Lutregala study area on South Bruny Island; C, sub-regional overview of the Wapengo study area; D, sub-regional overview of the Tarra study area; E, Lutregala salt marsh; F, Wapengo salt marsh; and G, Tarra salt marsh.White numbers correspond to transect numbers in Fig. 3. Blue transects were used as both surface and coring transects (Williams et al., 2021) and white transects were used only as coring transects (this study).Basemaps in E, F and G are from Google Earth.[Color figure can be viewed at wileyonlinelibrary.com]

Establishing chronologies
Dating of recent salt-marsh sediment can be problematic due to both the radiocarbon 'plateau' in the last 300 years, as well as the limited span (~100-150 years) of 210 Pb (Kemp et al., 2012).Therefore, to obtain a high-resolution chronology, we employed a range of dating techniques which included radiocarbon ( 14 C) dating by accelerator mass spectrometry (AMS), as well as 210 Pb, stable lead isotope ratios and pollen (cf.Gehrels et al., 2008Gehrels et al., , 2012;;Kemp et al., 2012).Samples for 14 C were analysed at the Scottish Universities Environmental Research Centre (SUERC), the University of California, Irvine (UCI) and the Australian Nuclear Science and Technology Organisation (ANSTO).We use bomb-spike AMS 14 C for sediments younger than ~60 years old and high-precision (multiple target) AMS 14 C (Marshall et al., 2007) or routine precision for sediments older than ~60 years old.We targeted horizontally bedded aboveground parts of terrestrial plants, but dated bulk sediment in several instances where insufficient macrofossils were encountered.Full details of the preparation methodology are given in supplementary file 1 (see Supporting Information S1).We calibrated pre-bomb spike 14 C dates using the SHCal20 calibration curve (Hogg et al., 2020) and calibrated post-bomb spike dates using the Bomb13 SH1-2 calibration curve (Hua et al., 2013) in 'rplum' (Aquino-López et al., 2020;Blaauw, 2021).
Preparation and analyses of the 210 Pb samples were undertaken at ANSTO by alpha particle spectrometry.Samples were processed for the determination of total 210 Pb, measured indirectly from its progeny polonium-210 ( 210 Po) and supported 210 Pb, measured from its grandparent radioisotope radium-226 ( 226 Ra).Unsupported 210 Pb activity was estimated by subtracting supported 210 Pb from the total 210 Pb activity.
Stable Pb isotope analyses were undertaken at the Geochronology and Tracers Facility at the British Geological Survey using a Nu Instruments, Nu Plasma HR, MC-ICP-MS (multi-collector-inductively coupled plasma-mass spectrometer).Sediment samples were taken at contiguous 1-cm intervals and prepared for isotope analysis (for the full preparation and mass spectrometry analysis methodology see S2 in Supporting Information 1).
Pollen and microcharcoal analyses were conducted at the University of Queensland to provide additional chronohorizons.Pollen preparation methods followed those outlined in Moss et al. (2013) and macrocharcoal preparation followed guidelines in Stevenson and Haberle (2005).
We generated Bayesian age-depth models in rplum to combine chronological information from the different methods and interpolate the age of every foraminiferal sample.As with traditional constant rate of supply models (e.g.Appleby & Oldfield, 1978), rplum assumes a constant rate of supply of 210 Pb; however, it separates the age-depth modelling process from the 210 Pb decay equation by using a self-adjusting Markov chain Monte Carlo (MCMC) algorithm (Sim et al., 2021).Calendar dates identified from chronohorizons in the core using the stable lead profile and pollen and charcoal records were assigned a date and uncertainty based on archival dates from the literature.

Reconstructing sea level
Following Gehrels et al. (2020), we used (error-in-variable) Gaussian process regression, accounting for both temporal and vertical uncertainties, to model sea-level change at Lutregala, Wapengo and Tarra.The temporal covariance (defining the variance of its mean and temporal correlations) was represented by a Matérn function with a smoothness factor of 3/2.
All corresponding hyperparameters (length scale for the definition of the temporal correlation between different points in time, and scale for the description of variability of the process) were estimated using an automatic hyperparameter optimization algorithm implemented in MATLAB's 'fitrgp' function.Both chronological and vertical uncertainties were assumed to be normally distributed.A Monte Carlo approach using 5000 iterations was used to estimate sea-level change at each site over time.Individual sea-level index points were glacio-isostatic adjustment (GIA) corrected using estimates from the Caron et al. (2018) model assuming that GIA is a linear process over the time periods considered here (~200 years).This model considers uncertainties by varying the rheology and ice history in each realization, and a representative set of 5000 GIA model iterations was used to obtain RSL estimates at each site (the 5000 estimates define the total size of the Monte Carlo experiment).As a result, 5000 different RSL reconstructions are derived from which median values and 95% confidence limits (2.5 and 97.5 th percentiles) were extracted.Compaction is considered negligible in the thin salt-marsh sediments from the region (Brain et al., 2012;Gehrels et al., 2012).
The selection of tide-gauge stations was based both on the geographical distance to the proxy sites and temporal coverage.Tide-gauge data were taken from Hogarth (2014) and GIA-corrected.Stations in close proximity to the marshes are limited to the last ~30 years.Long-term tide-gauge stations are located hundreds of kilometres from the sites and, in the case of Port Arthur, contain significant data gaps in the record (Fig. 2).
In the absence of long tide-gauge records in close vicinity to the marshes we follow Garrett et al. (2022) to calculate localized sea-level budgets (i.e.modelled estimates of the contributing components to sea-level rise) for each site.We use sterodynamic estimates from the SODAsi.3ocean re-analysis model (Giese et al., 2016), present-day barystatic (including GRD) estimates from Frederikse et al. (2020) and the inverse barometer effect calculated from sea-level pressure fields from the 20 th century re-analysis project v3 (Slivinski et al., 2019).GIA was not considered in the budgets as we compared them to GIA-corrected proxy and tide-gauge records.We do not consider any inherent uncertainties in the individual budget components, as they are only available for the present-day barystatic components, but not for the ocean reanalyses and the inverse barometer effect.
As proxy records filter interannual variability (i.e.several years up to several decades can be recorded in a single centimetre of a core depending on the sedimentation rate), tide-gauge records and budgets were 30-year smoothed.For the local tide-gauge data and budgets, this was done using a Singular Spectrum Analysis (Moore et al., 2005) with an embedding function of 15.For the long-term tide-gauge records, this was done using a Gaussian-weighted moving average filter as the records contain some significant data gaps and so Singular Spectrum Analysis could not be used.

Transfer functions
Transfer functions developed by Williams et al. (2021) were used to generate PMSE estimates at Lutregala and Wapengo (Table 1).For Tarra, a screened local training set (36 samples) of foraminifera was generated from the two transects, which, when combined, have a vertical range of 0.6 m (see Supporting Information Fig. S3.1).Species encountered are similar to those at Lutregala & Wapengo (Williams et al., 2021) including: Trochamminita salsa, Haplophragmoides wilberti, Polysacammina ipohalina, Trochammina irregularis, Entzia macrescens, Trochammina inflata, Siphotrochammina lobata, Miliammina fusca and Ammobaculites exiguus, as well as Ammotium fragile (not previously encountered at either site).There is little vertical zonation of foraminifera and T. inflata dominates across the majority of the vertical gradient (average 65% abundance, maximum 97% abundance).Detrended canonical correspondence analysis indicates a linear response of the foraminifera to elevation, and therefore a partial least squares regression model (Wold et al., 1984) was used.The resulting transfer function performed very poorly in terms of the relationship between observed and predicted elevations (root mean squared error of prediction: −0.08 m; r 2 boot : 0.02; see Fig. S3.2).Many of the samples were predicted at the same height, ~1.20-1.25 m (Fig. S3.2).We suggest this is due to the dominance of T. inflata across the marsh.As species optima are similar between foraminifera in the local and regional training sets (Fig. S3.3) there are several options: (i) use the existing southeastern Australia regional II training set for the Tarra reconstruction; (ii) include the Tarra samples in a regional training set with existing samples; or (iii) follow, for example, Long et al. (2014) and assign conservative semi-quantitative indicative ranges based on stratigraphy and common species.When samples are incorporated into the existing southeastern Australia regional transfer function (Williams et al., 2021), the model r 2 boot reduces from 0.69 to 0.54 and maximum and average bias increases as many of the samples, like in the local transfer function, are predicted at the same elevation.Furthermore, the Tarra surface samples do not provide a wider range of analogues than the existing regional training set; therefore, we do not include them.The present-day salt marsh extends between 1.35 and 1.01 m AHD, which yields an indicative range for salt-marsh sediments of 1.18 ± 0.17 m.We cannot assign a narrower vertical range based on comparing the fossil and local modern foraminiferal assemblages, as foraminifera found at both the upper and lower elevation limits in the modern environment at Tarra are present in the core.Using the regional II training set to estimate PMSE results in ranges of 1.12 ± 0.12 to 1.33 ± 0.12 m AHD, with an average of 1.20 ± 0.12 m AHD.Therefore, both approaches result in very similar outputs.Finally, we also compared the poorly performing local transfer function PMSE estimates with the regional PMSE estimates.The PMSE estimates of both models overlap significantly, further suggesting that the regional model is appropriate.

Lutregala stratigraphy and foraminiferal assemblages
We established two coring transects at Lutregala which followed the surface sampling transects (Williams et al., 2021;Fig. 1).At the base of the sequence at Lutregala (Unit 3; Fig. 3a) is a silty sand unit which extends down to at least −1.24 m AHD.Overlying this unit is a silt (Unit 2) which was thickest towards the seaward and landward sides of the salt marsh.At the surface of the sequence is a salt-marsh peat (Unit 1).The maximum peat thickness is 28 cm.Unit 1 was subdivided into a silty salt-marsh peat at the bottom (1b), and a fibrous salt-marsh peat at the top (1a).
The foraminiferal assemblages in the silty sand unit mainly consist of the high to mid-marsh species T. inflata (average relative abundance 96%), with a low abundance of T. salsa and E. macrescens.Total counts are low in the unit (average of 19 specimens per sample).Foraminifera in the silt unit largely comprise T. inflata (average 96% relative abundance) and average total counts are higher (average of 56 individuals per sample).Within the silty salt-marsh peat, unit 1b, T. salsa is more prevalent (average 39%, up to a maximum of 75% relative abundance).Into the fibrous salt-marsh peat, unit 1a, the relative abundance of T. salsa declines and the assemblages diversify, with T. inflata, E. macrescens and M. fusca increasing.Total counts are largest in this unit (average 219 individuals per sample).
Modern analogue results (Table 2) show that all samples have at least fair modern analogues in the regional or local Lutregala training sets, suggesting that fossil assemblages are mostly similar to modern assemblages.The larger number of fair rather than good analogues in the lower third of the core (Fig. 4a) is due to the relative abundances of T. inflata being higher in the core than in the modern surface samples (Williams et al., 2021).Whilst the regional model results in a larger number of good modern analogues compared to the local model, there is very little difference in PMSE estimates between the two models (Fig. 4a) and, owing to the reduced vertical error of the local model compared to the regional model, we choose to employ the local transfer function.
Only one sample from the silty sand unit contained more than 50 specimens; the resulting PMSE reconstruction indicates the sample accumulated at around 0.71 m AHD, an elevation corresponding to the modern mid-to high marsh.PMSE predictions in this silt show that the marsh elevation remained largely unchanged from the silty sand, and foraminifera remain indicative of a mid-to high salt-marsh environment.In the silty salt-marsh peat, corresponding with the increase in T. salsa, PMSE increased by 0.21 ± 0.08 m, reaching a maximum of 0.91 m AHD, showing an increase in the marsh elevation to a high-marsh environment.Into the fibrous salt-marsh peat, there was a fall in PMSE to 0.70 m AHD, which corresponds with the diversification of the assemblages to mid-to low marsh assemblages.In total, the local PMSE estimates range from 0.68 to 0.91 m AHD with an average uncertainty of ± 0.06 m (1σ).

Wapengo stratigraphy and foraminiferal assemblages
At Wapengo we established three coring transects, each from the landward to the seaward edge of the marsh (Fig. 1).The deepest core reached −1.03 m AHD and the basal unit is characterized by a silty sand (Unit 3).Overlying this unit is a silt (Unit 2).This unit is ~0.17 m in thickness and the stratigraphy is characteristic of a tidal-flat environment.Above the silt is a salt-marsh peat (Unit 1) which can be subdivided into a silty peat (Unit 1b) at the base and a fibrous salt-marsh peat at the top (Unit 1a).The area of thickest peat is noted from 0.73 to 0.97 m AHD (Fig. 3b).
The foraminiferal assemblage in the silty sand unit largely consists of T. inflata, E. macrescens and H. wilberti (averaging 69, 11 and 9% respectively), although total counts are low (average 26 individuals).The silt unit includes T. inflata and M. fusca; however, again, foraminiferal density is low (average of 15 individuals).Trochamminita salsa and T. inflata become dominant towards the top of the silt unit.The base of the silty peat contains T. salsa (average 23%, maximum 60%), although this decreases up the unit and is replaced by high counts of T. inflata (average 70%).Into the fibrous salt-marsh peat, the relative abundance of M. fusca and P. ipohalina increase and T. inflata remains abundant (averages of 33, 8 and 36% respectively).Modern analogue results (Fig. 4b; Table 2) indicate that the regional training set provides a larger number of good analogues for the Wapengo core than does the local training set, as well as no poor analogues.The local model results in 14 samples with poor analogues and fewer good analogues.This is due to the moderately high relative abundance of T. salsa in the fossil assemblages, but low abundance in the local modern training set.Trochamminita salsa was often found in low count samples at Wapengo near the limit of the highest occurrence of foraminifera or highest astronomical tide, and so were not included in the training set after screening by count size (Williams et al., 2021).However, the species is more abundant in the regional training set due to the higher abundance in high total count samples at Lutregala.Williams et al. (2021) compared the modern species optima between the Wapengo local model and the regional model.As optima were consistent between models, we employ the regional transfer function.
From the sandy silt to the silt, PMSE increases by 0.19 ± 0.11 m, reaching a maximum of 0.98 m AHD, elevations associated with high-marsh environments in the modern environment.PMSE reconstructions decline from the silty peat to the fibrous peat by 0.13 ± 0.11 m to reach the core-top value of 0.81 m AHD as lower marsh foraminifera become more dominant upcore.Overall, PMSE predictions from the regional transfer function range from 0.64 to 0.98 m AHD, with an average uncertainty of ± 0.08 m (1σ).

Tarra stratigraphy and foraminiferal assemblages
At Tarra we collected cores along four transects across the marsh from the landward to the riverine edge of the marsh (Fig. 1).The deepest core extends down to -0.41 m AHD into a yellow mottled silt (Unit 3; Fig. 3c).The mottled silt is overlain by a brown-grey silt containing a small proportion of red sand (Unit 2).The brown-grey silt is overlain by a silty peat unit (Unit 1b) which is overlain by an organic salt-marsh peat Table 2. Modern analogue technique results for each reconstruction using the local and regional transfer function models (Williams et al., 2021).Samples are classified according to the 5 th and 20 th percentiles of their minimum dissimilarity.unit (Unit 1a).The area of thickest peat is found from 0.91 to 1.08 m AHD.

Site
In the core, the yellow mottled silt unit contains very sparse foraminifera.The brown-grey silt unit largely contains T. salsa, T. inflata and M. fusca (average relative abundances 19, 51 and 15% respectively).Total counts of foraminifera in the unit were higher (average 163 individuals per sample).In the silty peat, T. salsa and M. fusca decreased and the relative abundance of typical mid-marsh species T. inflata and E. macrescens increased, with T. inflata becoming especially dominant (average relative abundance 68%).Into the fibrous organic salt-marsh peat, assemblages remained largely similar to the silty peat, although we note an increase in the relative abundance of E. macrescens (average relative abundance 25%).Modern analogue results (Fig. 4c; Table 2) show that the regional training set provides good analogues for 14 samples from the Tarra core and fair analogues for the remaining 13 samples.The majority of good analogues are found towards the top of the core where E. macrescens and T. inflata (common species in the modern training set) increase in abundance.No samples are classified as having a poor analogue, suggesting that the regional training set is representative of fossil material in the Tarra core.
PMSE could not be calculated for the yellow mottled silt unit of the core due to the low total counts.In the brown-grey silt, PMSE declined by 0.22 ± 0.18 m from 1.33 m AHD.In the peat units, PMSE remained largely uniform, showing the marsh kept pace with sea-level rise, with a small decline at the top of the core, reaching a core-top value of 1.16 m AHD.In total, PMSE predictions range from 1.12 to 1.33 m AHD with an average uncertainty of ± 0.12 m (1σ).

Chronologies
High-resolution sea-level reconstructions must be underpinned by robust and well-dated chronologies; therefore, we used a combination of radiocarbon 14 C AMS dating, radionuclide, stable lead ratios, as well as pollen and charcoal data to generate chronologies for each core.In the stable isotope Pb record, we identified chronohorizons including the onset of Southern Hemisphere pollution as a result of the mining, and subsequent smelting, of the Broken Hill lead-zinc-silver ore deposits (~1895 CE; Gehrels et al., 2008;Vallelonga et al., 2002).The Broken Hill smelting signal has been identified in several environmental archives across Australia (e.g.Gehrels et al., 2012;Kristensen et al., 2017;Marx et al., 2010;Wu et al., 2016).In each record, lead isotopes exhibit a distinct excursion to less radiogenic values, indicative of the lower isotopic signature of the ores compared to background (Broken hill ore: 206 Pb/ 207 Pb 1.04; 206 Pb/ 204 Pb ~16.00; 208 Pb/ 207 Pb 2.33; Chiaradia et al., 1997;Kristensen et al., 2017).A date of ~1890-1900 CE can be assigned to a chronohorizon where the isotopes diverge from the background rate (Gehrels et al., 2012).We also identify the Pb pollution maximum in Australia at ~1974 CE (following a peak in petrol consumption and subsequent Pb emissions; Kristensen, 2015).
At Lutregala, whilst we find exotic pollen indicative of the European settlement of Australia, other chronostratigraphic data from the core suggest that it did not appear until the midto late 20 th century; this may be attributable to the very low population density on South Bruny Island until the late 20 th century (Jackson, 2006).Similarly, no distinct horizon could be identified within the charcoal record.We obtained nine AMS 14 C dates from the master core (Table 3).After considering stratigraphic ordering and comparing calibrated dates to established chronohorizons, two dates were removed prior to age-depth modelling (UCIAMS-26610 and UCIAMS-236611).These dates were deemed too young for their stratigraphic position, possibly as a result of root contamination.Dates included in the model range from 1465 CE to modern.In total, we obtained 20 age estimates using this multi-dating approach and reject two (Fig. 5a; Supporting Information file 3).
Eleven samples from the Wapengo core were analysed for 210 Pb and we identified three chronohorizons from stable lead  isotopes and pollen (Fig. 5b).The stable isotope Pb record showed signatures of both the Broken Hill smelting as well as the Pb pollution maximum.The Pb pollution maximum was also corroborated by an increase in Pinus (pine) pollen at the same depth, presumably derived from pine plantations established in the 1960s-1980s (Keith & Bedward, 1999).We identified the European settlement layer from the introduction of exotic flora in the core at 21.5 cm.At the same depth, we also noted increases in Sporormiella (a dung fungus), which is associated with the presence of sheep and cattle in the catchment, as pastoralism in the region began (Lunney & Leary, 1988).We obtained eight AMS 14 C dates from the core (Table 3).However, a visual inspection based on stratigraphic ordering, and taking into account chronohorizons from the stable lead and pollen data, suggests that samples UCIAMS-236617, UCIAMS-236618 and UCIAMS-236619 have calibrated ages that are too young for their stratigraphic position and sample OZZ403 is too old for its stratigraphic position.Sample OZZ403 was a bulk sediment sample which may have contained reworked material, and the younger dates may have been contaminated by root penetration.As clear outliers, these dates were removed from the age-depth model.Accepted ages range from 638 CE to modern.In total we accept 18 age estimations from the core and reject four (Fig. 5b; Supporting Information file 3).Finally, eight samples were analysed for 210 Pb from the Tarra core and we identified three chronostratigraphic markers from stable lead isotopes and pollen (Fig. 5c) including the Broken Hill smelting, the Pb maximum and European settlement.Whilst exotic pollen taxa are noted at both 46.5 and 22.5 cm, the occurrences are isolated and do not increase significantly again until 15.5 cm.We therefore place European settlement at 15.5 cm based on the appearance and sustained increase of pine and Asteraceae (Liguliflorae, introduced European daisy) from this depth.Following Gehrels et al. (2008) we add a 20-year lag period with a ±10-year error to account for the time taken for the pine trees to produce pollen.We obtained a further eight AMS 14 C dates from the core (Table 3).A visual inspection using prior information from the identified chronohorizons and pollen data suggests that OZZ405, SUERC-93778 and SUERC-93779 are too old for their stratigraphic position.All three of the samples were bulk sediment and may have included reworked material.UCIAMS-236625 appears too young as it occurs at the same chronohorizon as the Broken Hill smelting and we place precedence on this marker over the 14 C date due to the clarity of the signal.We therefore removed these four dates from the age-depth model prior to analysis.Accepted dates range from 677 CE to modern.In total we accept 15 age estimates from the core and reject four (Fig. 5c; Supporting Information file 3).
Across all three sites, the age-depth models show low sedimentation rates in the sand and silt units at ~40-50 a cm −1 ; however, the rates increased substantially in the saltmarsh peats to <10 a cm −1 .Converting these sedimentation times to accumulation rates, in the sand and silt units, accumulation was 0.02-0.03cm a −1 (Fig. 4).This increased across all sites to 0.11-0.19cm a −1 in the salt-marsh peats, with Tarra having a lower accumulation rate compared to Wapengo and Lutregala (Fig. 4).
In all three models, the chronology is well constrained from the mid-19 th century, but age uncertainty is substantial before European settlement due to a lack of dateable organic material in the cores.Furthermore, foraminifera in the silt and silty sand units are incongruous with the facies and may reflect infaunal processes within the core.We therefore choose to base our sea-level reconstructions only on samples from the peat units where we have been able to obtain a high-resolution chronology and where foraminiferal assemblages are consistent with the sedimentary facies.This produced an RSL record from ca. 1830 CE to present.

Relative sea-level change in southeastern Australia
We reconstruct sea level over the last ~200 years at Lutregala, Wapengo and Tarra by combining RSL estimates with age estimations determined downcore at 1-cm resolution using the age-depth models.We also update the previously published Little Swanport reconstruction using the SHCal20 calibration curve (Hogg et al., 2020) and run the data through the same Gaussian process analysis.Corrections of −0.01 (−0.41 to 0.51; 95% confidence interval [CI]) mm a −1 , −0.14 (−0.52 to 0.28) mm a −1 , −0.25 (−1.16 to 0.30) mm a −1 and −0.05 (−0.70 to 0.46) mm a −1 was applied at Lutregala, Wapengo, Tarra and Little Swanport respectively, to account for GIA (Caron et al., 2018).
The rate of sea-level rise increased during the first half of the 20 th century  at all four sites to rates of 1.7 (−4.0 to 6.5; 95% CI) mm a −1 , 4.0 (1.2 to 7.1) mm a −1 , 1.4 (−2.0 to 4.7) mm a −1 and 4.0 (1.6 to 7.1) mm a −1 at Lutregala, Wapengo, Tarra and Little Swanport respectively (Fig. 7).The acceleration is especially pronounced at Wapengo and Little Swanport, although the onset of more rapid sea-level rise occurs at different times (Fig. 6).At Tarra, the reconstruction does not show evidence of a pronounced acceleration in sea level, but rather a gradual continuous rise over the 19 th and 20 th centuries.
Whilst maximum rates of 8.3 (4.4-16.8;95% CI) mm a −1 , 5.4 (2.1-8.9)mm a −1 and 5.4 (3.0-8.4)mm a −1 are reconstructed at Lutregala, Wapengo and Little Swanport during the early to mid-20 th century, maximum rates were not reached until the early 21 st century at Tarra (Fig. 7).From the mid-20 th century to present, the records from Lutregala, Wapengo and Little Swanport largely converge with Tarra and suggest a slowing in the rate of sea-level rise, although, due to the exceptionally high rates in the early 1950s, average rates are higher at Lutregala compared to the other sites.Our reconstructions suggest considerable regional variability with average rates of 4.6 (0.9 to 8.7) mm a −1 , 0.9 (−2.1 to 2.8) mm a −1 , 1.2 (−2.5 to 4.3) mm a −1 and 0.4 (−2.6 to 3.3) mm a −1 from 1950 to 1999 at Lutregala, Wapengo, Tarra and Little Swanport respectively.

Synthesis of proxy and tide-gauge records
Our three new proxy records from Wapengo, Lutregala and Tarra, combined with the revised record from Little Swanport, provide a regional view of sea-level change over the last 200 years, supplementing and extending the sparse and temporally limited tide-gauge network.A comparison with tidegauge data over the last three decades shows good agreement with our reconstructions at Lutregala and Tarra.Tide-gauge data from Stony Point, in particular, agree well with the Tarra proxy record, falling largely within the Gaussian process range.Tide-gauge data at Eden, the closest station to Wapengo, predict marginally faster rates of sea-level rise than the Gaussian process range but are within the vertical uncertainty of the data.There is little temporal overlap between available instrumental data from Spring Bay and the Little Swanport proxy record, although where the data do overlap, the observational data fall within the Gaussian process range (Fig. 6).
The oldest documented tidal measurements in Australia come from Port Arthur, Tasmania.However, the tide gauge only recorded sea level continuously between 1841-1842and 1999-2002(Hunter et al., 2003)).Sporadic measurements were also taken between 1875 and 1905 and in 1972.The earliest measurements from Port Arthur suggest sea level was 0.14 ± 0.04 m (2σ below present (expressed relative to 2002) in ~1841-1842~1841- (cf. Gehrels et al., 2012)).Only the Wapengo and Little Swanport records are long enough for comparison to the Port Arthur data, but the Gaussian process indicates that sea level was ~0.27 ± 0.32 and ~0.32 ± 0.45 m below present levels ( 2002) respectively in 1841.This indicates a discrepancy of ~0.1-0.2 m between the proxy records and the Port Arthur data for this time period.Differences cannot be related to GIA given the similarity in estimates of GIA for Tasmania and mainland Australia and the timescale of the records, but may relate to significant uncertainty surrounding the installation of the benchmark at Port Arthur (Hunter et al., 2003), or lack of high-resolution temporal data from the tide gauge.For example, some annual data in the Port Arthur record are based on a single measurement (Hunter et al., 2003).Linear regression of the tide-gauge record indicates a GIA-corrected trend of 1.0 ± 0.6 mm a −1 (2σ between 1841 and 2002 (Hunter et al., 2003).This is slower than both the proxy records suggest, and also slower than other long-term tide gauges from the mainland (e.g.Gehrels et al., 2012;Hogarth, 2014;White et al., 2014;Hague et al., 2021).Comparison of the trend derived from the Port Arthur tide gauge to sea-level trends from neighbouring tide gauges is not possible due to the lack of long-term data from Tasmania, and so it remains difficult to assess the veracity of the record.
When compared to the two closest near-continuous longterm tide-gauge records from the region, Fort Denison and Williamstown, only Tarra demonstrates similar magnitude rises in sea level to present (~0.20 m), with the proxy records generally suggesting larger rises in sea level than the instrumental data (~0.30-0.35m).The few long-term tide gauges in southeastern Australia also have data gaps, subsidence issues and datum shifts, especially over the late 19 th and early 20 th century, which can make them problematic (e.g.Hogarth, 2014;White et al., 2014;Hague et al., 2021).Hogarth (2014) suggests that the Williamstown and Port Arthur records, in particular, should be analysed with caution.Recent efforts by Hague et al. (2021) have made corrections to the Williamstown record since 1966, and there are ongoing efforts to digitize and correct this record prior to this (B.Hague, 2022, personal communication).Similarly, in the longest record, Fort Denison, there were also issues with digitization of data, which may have led to errors in the earlier part of the record (Hamon, 1988;White et al., 2014).Hague et al. (2021) state that high-quality tide-gauge data are only available from 1966 onwards after the systematic collection of data commenced across Australia.

Timing of the acceleration
The ability to detect significant accelerations in sea level in instrumental records can be difficult owing to large inter-annual variations resulting from localized atmospheric and oceanic processes, as well as the necessity for records to be long enough (i.e.>40 years) to capture accelerations (e.g.Haigh et al., 2014;Steffelbauer et al., 2022).White et al. (2014) state that individual Australian tide-gauge records are too short and too variable for the detection of statistically significant accelerations in sea-level rise.Our proxy records therefore provide an alternative means for determining the timing of accelerations, albeit at a much lower temporal resolution than tide-gauge records.Average rates of sealevel change exceeded 0 mm a −1 in ~1833-1918 (95% CI), ~1871-2018 (95% CI), ~1821-1885 (95% CI) and ~1905-1945 (95% CI) at Wapengo, Tarra, Little Swanport and Lutregala respectively (Supporting Information Fig. S14).There is therefore significant inter-site variability in the timing of the onset of acceleration.Even when accounting for GIA uncertainty at the 95% confidence interval, there is not one common acceleration period between all of the reconstructions, suggesting other causes underlying the variability (Fig. S14).

Drivers of 20 th century sea-level change
At all sites, the sum of our best estimates of sterodynamic, barystatic GRD, and inverse barometer effect contributions largely falls within the uncertainty of the reconstructions during the first half of the 20 th century and generally lies within the Gaussian process ranges from ~1950 to present, suggesting that known contributors can largely explain the observed sea-level rise at our sites (Fig. 7).A discrepancy between the budget and the Wapengo reconstruction occurs between 1900 and 1925 when there is an increase in M. fusca in the foraminiferal population.This discrepancy may relate to infaunality of M. fusca, as the budget highlights a rise in sea level immediately following this increase.The largest discrepancy between the modelled budget and the proxy data is apparent in the Little Swanport reconstruction between ~1950 and 1970.This occurs where relative abundances of T. irregularis are highest.This resulted in high PMSE estimates in the reconstruction ~0.45-0.75m AHD (Gehrels et al., 2012) but is contrasted by a rise in sea level in the sea-level budget.The discovery of a new foraminiferal genus and species in Australia and New Zealand Pseudotrochamminita malcomi (previously identified as either T. salsa or T. irregularis; King, 2021) indicates that P. malcomi has, overall, a lower elevation optimum than T. salsa and T. irregularis.Therefore, we suggest the PMSE estimates in this section of the core may be too high owing to the potential combining of two species into one.The discrepancy could also relate to uncertainties in the age-depth model, though we note that this section of the core is well dated by numerous chronohorizons (Gehrels et al., 2012), with the post-1975 budget aligning with the median estimates of the reconstruction.
The budget indicates that both the barystatic GRD and sterodynamic components have contributed to the rapid rates of sea-level rise.The barystatic GRD contribution was high over the first half of the 20 th century.Overall, it contributed an average of 1.6 mm a −1 to sea level at our sites between 1900 and 1949, with greater contributions of ~2 mm a −1 especially between 1920 and 1940 (Fig. 8).The barystatic component also dominated early 20 th century global mean sea-level rise (e.g.Parkes and Marzeion, 2018;Frederikse et al., 2020;Malles & Marzeion, 2021), with contributions primarily from land-based glaciers, as well as the Greenland and Antarctic Ice Sheets (Frederikse et al., 2020).
The sterodynamic component, as modelled in SODA, then amplified the rapid rates and subsequently drove the acceleration from ~1930 to 1940, contributing up to ~4 mm a −1 at its peak in the mid-20 th century (Fig. 8).The increase in sterodynamic sea level may have been driven by strengthening of ocean gyres (including the East Australian Current; Cai and Cowan, 2007;Giese et al., 2016)  temperature (especially from ~1920), changes in wind stress (Giese et al., 2016) and phase changes of several key climate modes.For example, over the early 20 th century, the Interdecadal Pacific Oscillation (IPO) index declined (e.g.Salinger et al., 2001;Parker et al., 2007), changing from a positive to a negative phase change.The IPO modulates the impacts of the El Niño Southern Oscillation (ENSO) in Australia by controlling the position of the South Pacific Convergence zone and influencing sea surface temperature (and thus sea level) (e.g.Power et al., 1999;Henley et al., 2015;Kelly et al., 2019).This rapid decline in the IPO (marked by a positive increase in the rate of change of the index) is coincident with our rapid rates of sea-level rise.Additionally, an upward trend of the Southern Annular Mode (SAM) due to the depletion of ozone has resulted in higher rates of sea-level rise in the subtropical southern oceans over the late 20 th century and early 21 st century (Duan et al., 2021).
It is important to note that sterodynamic changes are the only component that show significant variability between the locations of our salt-marsh records.Variability is especially large over the first half of the 20 th century, after which sterodynamic contributions become more comparable between sites (Fig. 8).As mentioned, this is largely replicated also in the proxy data.However, variability between modelled sterodynamic contributions may also be due to larger uncertainties in the ocean reanalysis before 1950 due to, for example, drifts resulting from less assimilated data or uncertainties in heat and freshwater fluxes (Storto et al., 2019).Even for the second half of the 20 th century, ocean reanalyses often show multi-millimetre differences in linear trends of sterodynamic sea level (Dangendorf et al., 2021).Those differences become larger further back in time.Future work may seek to utilize an ensemble of sterodynamic estimates to investigate whether uncertainties in budget estimates may explain discrepancies with proxy reconstructions.
A common feature of the Lutregala, Wapengo and Little Swanport records, as well as the New Zealand proxy records (Gehrels et al., 2008;Garrett et al., 2022), is a fall or plateauing of sea level from the mid-20 th century to the late 20 th century.This trend has also been found in tide-gauge records around Australia and New Zealand.For example, White et al. (2014) noted a sea-level depression of ~0.2 mm a −1 for the majority of the east Australian coast from 1966 to 2010.Our records from Wapengo and Lutregala show a depression with average rates of ~0.0-0.1 mm a −1 over the same period.Similarly, Haigh et al. (2011) also observed a period of stability in tide-gauge records from Western Australia between 1950 and 1990, with a rise in the early 21 st century.The stabilization in sea level in the western Pacific over the latter half of the 20 th century has also been attributed to phase changes and an intensification of climate modes (e.g.Goring and Bell, 1999;Church et al., 2004;Sasaki et al., 2008;Woodworth et al., 2009;Haigh et al., 2011;Holbrook et al., 2011;Watson, 2011;White et al., 2014).An El Niño (negative ENSO) phase depressed sea level around Australia (Goring & Bell, 1999) and the index became progressively more negative after ~1960, resulting in more El Niño phases.Our (sterodynamic) budget reconstruction estimates a decline in the rate from ~1960 to 2000 coinciding with a trend of more frequent high-pressure events over Australia (Figs. 7 and 8).However, the plateau is not synchronous in time amongst the reconstructions, and therefore warrants further investigation via addition of new reconstructions in the region, and by undertaking a spatiotemporal analysis of regional sea-level trends to better decipher the regional and local non-linear and linear signals in the RSL records (cf.Walker et al., 2022).
Finally, it is possible that foraminifera respond to non-tidal (i.e.wave and wind-induced) changes in water level (Woodroffe & Long, 2010).Kemp et al. (2022) state that non-tidal water-level variability can bias reconstructions, particularly those from microtidal regimes.If non-tidal variability is relatively large in comparison to tidal variability, the relationship between elevation and inundation can be distorted.Along the US East Coast, at some sites with a tidal range <1.0 m, non-tidal forcing drove water levels above Highest Astronomical Tide.In this study, efforts have been made to accurately capture the upper limit of marine influence (i.e.inundation frequency) by using the highest occurrence of foraminifera (Wright et al., 2011;Williams et al., 2021) rather than a predicted astronomical datum.This limits bias in the accuracy and precision of the vertical uncertainty of our palaeomarsh estimates, and results in more robust reconstructions.We assume that the inundation regime has remained stationary over time, but suggest that any bias resulting from this assumption would be negligible given the centennial timescale of the study.There is also a possibility of issues of re-working and infaunality (especially in the Little Swanport reconstruction, for which the earliest sea-level index points were derived from tidal-flat deposits).This is supported by the low accumulation rate of the tidal-flat units, low preservation of foraminifera in the units and age reversals in the chronology.Adding further high-resolution longterm records in the region will help to resolve discrepancies and further test hypotheses regarding driving mechanisms of sea-level changes in the Tasman Sea.

Conclusions
We have generated three new RSL reconstructions for southeastern Australia (Tasmania, Victoria and New South Wales) using salt-marsh foraminifera spanning the period ca.1830-2018.We also updated a previously published record from Little Swanport, Tasmania (Gehrels et al., 2012), using the SHCal20 calibration curve (Hogg et al., 2020).Records from both Lutregala and Wapengo show a sea-level acceleration commencing either in the late 19 th (Wapengo) or early 20 th (Lutregala) century, which is consistent with the previous reconstruction from Little Swanport which exhibited an acceleration in the late 19 th century, reaching maximum rates of sea-level rise in the early 20 th century.The record from Tarra (Victoria), however, does not indicate a significant sealevel acceleration, but shows a gradual continuous rise in sea level to present, which is more consistent with nearby tidegauge data.Overall, our records suggest sea level has risen by 0.2-0.3m over the last 200 years.Similarly to previous findings from southeastern Australia and New Zealand (Gehrels et al., 2008(Gehrels et al., , 2012)), rates of sea-level rise in the first half of the 20 th century are rapid, with average rates of up to 4.0 (−0.4 to 7.1) mm a −1 between 1900 and 1949.
Comparisons between tide gauge and proxy data are hindered by the lack of long-term observational data from the region.To address this issue, we compare our new records to a modelled sea-level budget for each site based on sterodynamic sea level from an ocean reanalysis model, corrected for the inverse barometer effect, and the GRD fingerprints of barystatic sea level.Comparisons between the sea-level budget and the proxy records show that the budgets are largely within the reconstruction uncertainty at all sites, and the combined sterodynamic and barystatic GRD contributions can account for the observed RSL at our sites.Future work may also seek to compare reconstructions to modelled sea-level budget estimates rather than (or alongside) tide-gauge records in regions where there is limited temporal and spatial availability of reliable tide-gauge data, such as in the Southern Hemisphere.Such work should also involve multiple ocean reanalysis products with ensemble uncertainties (Piecuch et al., 2016).
Increases in sterodynamic components (probably linked to changes in equatorial surface wind stress) have driven, in part, the rapid rates of sea-level rise in the western Pacific over the early to mid-20 th century (Cai & Cowan, 2007;Giese et al., 2016).However, the rates of sea-level rise witnessed in our records are faster than those suggested from comparative sea-level records from the Northern Hemisphere (e.g.Gehrels et al., 2005;Leorri et al., 2008;Kemp et al., 2011), which raises the possibility that sea level in Australia may have been additionally driven by barystatic rise as a result of enhanced Northern Hemisphere ice melt during the early 20 th century.To test this more robustly, the individual contributions from the Greenland and Antarctic Ice Sheets and land-based glaciers need to be further constrained.
Our records indicate that there is significant regional variability between the reconstructions over the first half of the 20 th century, with differences in both the onset and pattern of sea-level rise.When accounting for GIA uncertainty, there is not one common acceleration period between all four reconstructions.We therefore suggest that variability may be attributed to changes in sterodynamics in the first half of the 20 th century as barystatic GRD effects and inverse barometer contributions remained highly coherent between sites.However, it is possible that localized factors could be driving the variability, including foraminifera responding to wind-or wave-induced changes in water level.Adding more highresolution proxy records and undertaking detailed spatiotemporal analyses by collating the existing and new records to isolate the local and regional non-linear signals may help elucidate sources of variability.Nonetheless, the records presented here make significant progress towards understanding early 20 th century spatiotemporal patterns of RSL change in southeastern Australia.They supplement and extend the existing tide-gauge network and can be included in global analyses to address, in part, the lack of long-term sea-level data from the Southern Hemisphere.
Age data for Lutregala, Wapengo and Tarra (excel file)https://doi.org/10.6084/m9.figshare.17129597 All other data are available in the Supporting Information of this article.

Supporting information
Additional supporting information can be found in the online version of this article.

Figure 2 .
Figure 2. Locations of the tide gauges closest to each site (coloured dots) and locations of long-term tide gauges (grey dots).Horizontal bars are scaled to the length of the tide-gauge record in yearsan example is given in the key.The Port Arthur tide gauge location is shown, but the record length is not plotted owing to significant data gaps in the record.Marshes at Wapengo, Lutregala and Tarra (this study) and Little Swanport (Gehrels et al., 2012) are represented by white squares.[Color figure can be viewed at wileyonlinelibrary.com]

Figure 3 .
Figure 3. Stratigraphy of selected coring transects at A, Lutregala; B, Wapengo; and C, Tarra showing location of master cores used for foraminiferal and chronological analyses.[Color figure can be viewed at wileyonlinelibrary.com]

Figure 5 .
Figure 5. Core chronologies for A, Lutregala; B, Wapengo; and C, Tarra showing full 2σcalibrated 14 C ranges (see Table 2 for the multiple possible ranges), the identified chronohorizons from stable lead and pollen as well as the detected unsupported 210 Pb in the cores.The resulting age-depth model for each core is also shownthe light grey ribbon represents the 2σ age uncertainty and dotted line indicates the mean age.See Supporting Information S11-13 for full models.[Color figure can be viewed at wileyonlinelibrary.com]

Figure 6 .
Figure 6.Comparison of GIA-corrected relative sea-level (GCSL) records from Wapengo, Tarra, Little Swanport and Lutregala showing the Gaussian process regression median and uncertainty (95% confidence intervals).Also shown are the relative sea-level estimates (grey boxes) given to 2σ.Long-term tide-gauge data (Gaussian-weighted moving average 30-year smoothed), the local modelled sea-level budget and local tide gauges (Singular Spectrum Analysisembedding function of 15; Moore et al., 2005) are also plotted.Local tide gauges for each site are: Eden (Wapengo), Stony Point (Tarra), Spring Bay (Little Swanport), Hobart (Lutregala).Sites shown trending N-S.[Color figure can be viewed at wileyonlinelibrary.com]

Figure
Figure Comparison of the rate of relative sea-level change (95% confidence interval) at the salt marshes and the full sea-level budget (black lines, Singular Spectrum Analysisembedding function of 15; Moore et al., 2005) from 1900 to 2020.Sites shown trending N-S.[Color figure can be viewed at wileyonlinelibrary.com]

Fig. S3. 1 .
Distribution of surface foraminifera at Tarra salt marsh.Fig. S3.2.Tarra local transfer function performance.Fig. S3.3.Species optima of the Tarra local training set and the Regional II training set.

Table 1 .
Summary of transfer function performance statistics.
RMSEP, root mean squared error of prediction; AHD, Australian Height Datum; SWLI, standardized water level index.SEA-LEVEL CHANGES IN SOUTHEASTERN AUSTRALIA

Table 3 .
Calibrated radiocarbon dates from plant macrofossils and bulk sediment for the Lutregala, Wapengo and Tarra cores.Samples that were removed from the age-depth model prior to analysis due to their unreliable age based on their stratigraphic position are highlighted ( †).Bomb-spike dates have multiple possible age ranges (see calibrated radiocarbon age column); the most likely is selected by rplum in age-depth modelling.
, increases in sea-surface © 2023 The Authors Journal of Quaternary Science Published by John Wiley & Sons Ltd.