Watson, P.J., 2020. Updated mean sea-level analysis: Australia. Journal of Coastal Research, 36(5), 915–931. Coconut Creek (Florida), ISSN 0749-0208.
As an island nation with 60,000 km of open coastline and extensive margins of increasingly urbanised intertidal estuarine foreshores, Australia is critically exposed to the global threat posed by rising sea levels into the future. This study provides a contemporary assessment of sea-level rise around Australia to the end of 2018, based on all available tide gauge records and satellite altimetry. The study provides the first national assessment of vertical land motion (VLM) around the coast, identifying margins more prevalent to subsidence, which in turn exacerbate the localised effects of a rising global mean sea level. These areas include coastlines between Townsville and Coffs Harbour, Burnie to Port Pirie, and Fremantle to Wyndham. State-of-the-art time-series analysis techniques applied to all high-quality tide gauge records exceeding 75 years in length (four sites) enabled improved insights into the temporal resolution of current rates of rise and accelerations in mean sea level around Australia than were previously available. Averaged across these four records in 2018, approximately 40% of the “relative” velocity observed (∼2.2 ± 1.8 mm/y, 95% confidence limit [CL]) is attributable to VLM. When corrected for VLM, only the Fort Denison site exhibits “geocentric” mean sea-level velocity in 2018 exceeding 2 mm/y. The average geocentric velocity across all four sites in 2018 equates to 1.3 ± 2.0 mm/y (95% CL). Interestingly, each long record exhibits similar temporal characteristics, whereby a low point in the velocity time series occurs sometime in the period from 1970 to 1990, after which velocity increases over time to a peak occurring sometime after ca. 2010, suggesting the presence of a small acceleration (albeit not statistically different to zero at the 95% CL) in the record.
Climate change is predicted to have far-reaching physical, social, environmental, and economic impacts (e.g., Houser et al., 2015; IPCC, 2014; Melillo, Richmond, and Yohe, 2014; Neumann et al., 2015; Watkiss, 2011). The capacity for mankind to adapt will (in part) be governed by the pace at which impacts will manifest and the success of global adaptation endeavours that might offset (or delay) the inevitability of impacts from longer-term climate change processes associated with radiative forcing mechanisms, such as sea-level rise (Watson, 2018).
The threats to human populations from current and projected climate change–induced rise in mean sea levels are profound (IPCC, 2014). As an island nation with some 60,000 km of open coastline (Geoscience, 2019), coupled with more extensive margins of intertidal estuarine foreshores, Australia is critically exposed to the global threat posed by sea-level rise. This threat is exacerbated by the so-called “suburbanisation” of the coast, which is now widely recognised as an emerging, if not already emerged, feature of Australia (Smith and Doherty, 2006). Population trends indicate that over the last 100 years, the majority of Australia's population has shifted from rural inland Australia to the capital cities, with a more recent shift towards the suburbanisation of the coast (Smith and Doherty, 2006), with some 85% of the population living within 50 km of the coast (Australian Government, 2009).
More than $226 billion AUD in commercial, industrial, road, rail, and residential assets are estimated to be at risk around Australia from sea-level rise alone by 2100, if greenhouse gas emissions continue at high levels (Steffen et al., 2019).
The prominence of the climate change issue has placed more emphasis on examination of the extensive global repository of “relative” (that is, relative to the land) mean sea-level records (Holgate et al., 2012), which, along with temperature and carbon dioxide, remain the key proxy data sets used to monitor and quantify changes in the global climate system (Watson, 2016a).
This study updates previous Australian sea-level assessments (e.g., Watson, 2011; White et al., 2014), providing improved temporal resolution in the mean sea-level signal through the use of more advanced time-series analysis techniques tailored specifically for sea-level research (Watson, 2018, 2019a). In turn, the analysis provides better instruction on the associated time-varying velocity and accelerations in each of the four records exceeding 75 years in length, highlighting key spatial characteristics of mean sea-level rise around mainland Australia at the end of 2018. In addition, this research was augmented by satellite altimetry available post-1993 to assess sea surface height (SSH) trends within the bordering sea margins around the continent and to estimate vertical land motion (VLM) around the coastline (Watson, 2019b), highlighting areas where the threat from sea-level rise is likely to be exacerbated.
Data Sources Used in This Study
Annual and monthly average time-series data from the public archives of the Permanent Service for Mean Sea Level (PSMSL) were used in the analysis, up to and including 2018 (Holgate et al., 2012; PSMSL, 2019). As a minimum, each site was required to contain monthly average data spanning the period from January 1993 to December 2018 to estimate VLM (see “Methods”). Only annual average time series longer than 75 years (minimum coverage 1943–2018) were considered for trend analysis. It is recommended to use data with a minimum length of ∼60–80 years for analysis of trends and acceleration in mean sea-level records (e.g., Douglas, 2001; Houston and Dean, 2013; Watson, 2018). This permits the optimum ability to separate low-frequency natural oscillations from the underlying trend (refer to “Methods” and “Discussion” for full details).
Critical additional conditions were imposed on the input data for trend analysis, specifically regarding the extent of data gaps (where relevant), in order to maintain the integrity of records and not thus unduly affect trend determination. These conditions included limiting the extent of gaps to 15% of the record, with a maximum continuous gap limited to 5% of the record length (Watson, 2018).
Some 58 sites met the aforementioned conditions for VLM assessment. Five sites met the data requirements for trend analysis (Fremantle, Fort Denison, Newcastle, Port Adelaide, and Port Pirie). However, following analysis, it became evident that there are as-yet-unidentified issues relating to the suitability of the Port Pirie record after the mid-1990s (refer to “Discussion” for further details). As a result, the Port Pirie record was not considered further for trend analysis. The 58 sites analysed were notionally assigned a station ID commencing with Darwin (ID = 1) in the Northern Territory and progressing clockwise around the country to Wyndham (ID = 58) in Western Australia (Figure 1; Table 1).
Satellite altimeter products supplied by the European Commission's Copernicus Marine Environment Monitoring Service (CMEMS, 2019) were used to extract time series of SSH anomalies. The two-satellite merged global gridded L4 (008_057) product was made available for this research in netCDF format with daily outputs spanning the period 1 January 1993 to 13 January 2019 on a spatial resolution grid of 0.25° × 0.25° (Cartesian). This data set was used to estimate VLM at each tide gauge site based on trends from differenced altimetry–tide gauge techniques (ALT-TG; refer to “Methods” for further details).
Direct VLM estimates from Global Navigation Satellite Systems (GNSS) were available from the Nevada Geodetic Laboratory (NGL, 2019), for comparison to the ALT-TG estimates. GNSS estimates were used only where the GNSS station was within 10 km of the tide gauge, and the record was at least 10 years in length and still operational.
Similarly, gridded SSH trends for the period from September 1992 to May 2019 were made available in netCDF for each of the sea margins around Australia from multimission Ssalto/Duacs altimetry data from CMEMS, distributed by Archiving, Validation and Interpretation of Satellite Oceanographic (AVISO, 2019). Trends were based on simple linear regression analysis and provided on the same spatial resolution grid of 0.25° × 0.25° (Cartesian). These data were not adjusted for glacial isostatic adjustment (GIA). Associated error margins were not provided with the AVISO gridded trends.
Differing methodologies were applied in this study, which can be appropriately partitioned into trend analysis of tide gauge records and estimates of VLM. All analysis and graphical outputs were developed by the author from customized scripting code within the framework of the R Project for Statistical Computing (R Core Team, 2019).
Trend Analysis of Tide Gauge Records
The complexity of the influences embedded within conventional monthly and annual average ocean water-level data sets has led sea-level research toward successively more sophisticated time-series analytical techniques. The key requisite for sea-level researchers remains isolating the comparatively small, nonstationary, nonlinear mean sea-level signal from the significant and substantial dynamic interdecadal (and other) influences and noise. This is achieved through the application of singular spectrum analysis (SSA) techniques adapted specifically for mean sea-level research (Watson, 2018).
SSA has proven to be an optimal analytic method for this task in sea-level studies (Watson, 2016b) because it is a powerful data-adaptive technique capable of decomposing a time series into the sum of interpretable components with no a priori information about the time-series structure (Alexandrov et al., 2012; Golyandina and Zhigljavsky, 2013). Specifically, SSA can efficiently decompose an original record into a series of components of slowly varying trend, oscillatory components with variable amplitude, and a structureless noise (Golyandina, Nekrutkin, and Zhigljavsky, 2001).
Only the longest annual average Australian records available from the PSMSL (>75 years) were considered for trend analysis. The general methodology applied in analysing the observational tide gauge records has been well established in the recent literature (Watson, 2016b, 2016c, 2018) and applied to the longest records available around the United States (Watson, 2016a), Europe (Watson, 2017), and South Korea (Watson, 2019a). This methodology has been slightly updated and refined to accommodate the recent published findings of Mann, Steinman, and Miller (2020). The procedure can be broadly summarised in the following five steps, facilitated via the R extension package “TrendSLR” (R Core Team, 2019; Watson, 2019c):
Step 1: Gap-Filling of Time Series
This is a necessity in order to decompose the time series using SSA. Whilst the longest and most complete tide gauge records have been used for analysis, missing data persist in several records. Where required, records have been filled using an iterative SSA procedure (Kondrashov and Ghil, 2006) in the first instance, which has an (assumed) advantage in preserving the principal spectral structures of the complete portions of the original data set in filling the gaps.
Step 2: Estimation of “Relative” Mean Sea Level
Having necessarily filled the time series in step 1, the record is then decomposed using one-dimensional SSA to isolate components of a slowly varying trend (i.e. mean sea-level change resulting from a combination of external climate forcing and anthropogenic greenhouse gas emissions) from oscillatory components with variable amplitude, and noise. From the SSA decomposition, “relative” mean sea level is estimated by summation of “trend-like” components. Thus, understanding the transition point between the trend (of primary interest) and low-frequency natural oscillations becomes the key consideration. From the SSA decomposition, the components are ordered based on their contribution to the original time series, deeming that the first component must constitute the trend. With the use of long annual average mean sea-level time series, the key issue is to identify any other components with “trend-like” characteristics.
One method of quantitatively considering such characteristics is to apply a periodogram to each of the other components from the SSA decomposition to inspect their spectral properties and apply techniques like frequency thresholding (Alexandrov and Golyandina, 2005). From an analysis of the literature, in particular, the recent published works of Mann, Steinman, and Miller (2020), “trend-like” components have been distinguished on the basis of having a relative contribution generally above 0.5 in the low-frequency range [0–0.02]. Similarly, such components commonly comprise the peak and second highest spectral densities confined to the lowest two frequency bands, in either order. These principles can be used to quantitatively isolate components most likely to represent the trend in mean sea level (refer to the “Discussion” for more details).
Whilst these general conditions help to rapidly isolate the components of interest, visual inspection, diagnostic assessment, and professional judgement are also still applied to confirm the composite of the low-frequency “trend” of primary interest.
Step 3: Estimation of “Relative” Mean Sea-Level Velocity and Acceleration
The “relative” mean sea-level velocity and acceleration can be readily estimated from the first and second derivatives, respectively, of a cubic smoothing spline fitted to the “relative” mean sea level (or trend) determined in step 2. In each case, a fitted spline (with approximately 1 degree of freedom for every 10 years of record) results in the coefficient of determination (R2) of the fitted spline to the sea-level trend exceeding 0.99, providing a high degree of confidence in this form of model to estimate the associated time-varying velocity and acceleration.
Step 4: Estimation of Errors
This process initially involves fitting an autoregressive time-series model to remove the serial correlation in the residuals between the SSA-derived trend (step 2) and the gap-filled time series (step 1). The estimation of error in the “relative” mean sea level and associated velocity and acceleration is then based on bootstrapping techniques, where the uncorrelated residuals are randomly recycled, and the processes in steps 2 and 3 are repeated 10,000 times. From the extensive pool of outputted “relative” mean sea level and associated velocities and accelerations, standard deviations are readily calculated to derive robust confidence intervals.
Step 5: Correction to Estimate “Geocentric” Velocity
The correction from “relative” to “geocentric” velocity (that is, to a fixed datum) is based on VLM estimated from the approach advised in the following section. Error margins in quantifying “geocentric” velocity are determined in quadrature.
Following the approach advocated by Ostanciaux et al. (2012) and further updated by Watson (2019b), VLM was estimated using an ALT-TG technique by applying a least squares linear regression fit to the difference between the monthly averages derived from satellite altimetry and the “relative” tide gauge record at a point of interest. More specifically, to estimate VLM at a specific tide gauge site, the methodology applied can be broadly summarised in the following three steps:
Step 1: Determine nearest CMEMS Grid Reference with Data
Watson (2019b) found that ALT-TG VLM estimates are substantially improved by using gridded altimetry SSH anomaly products no closer than 30 km from the open coast and islands. While satellite altimeters perform very well over the open ocean, several issues arise in the vicinity of land, related to poorer geophysical corrections and artefacts in the altimeter reflected signals linked to the presence of land within the instrument footprint (Gommenginger et al., 2011). The available CMEMS satellite altimetry data were provided on a spatial resolution grid of 0.25° × 0.25°, and so the initial step was to isolate the nearest grid reference point to the tide gauge location that was no closer than 30 km from the open coast and islands and that contained complete SSH data.
Step 2: Convert Daily CMEMS SSH Time Series to Monthly Averages
Daily CMEMS SSH anomaly data spanning 1 January 1993 to 13 January 2019 were converted into monthly average time series spanning the period from January 1993 to December 2018.
Step 3: Linear Regression Analysis
VLM can be estimated from linear regression of the difference between the respective monthly average value of the CMEMS altimetry (step 2) and the tide gauge time series spanning January 1993 to December 2018. VLM was estimated as a rate in millimetres per year with associated standard errors. The estimated VLM was then used to correct “relative” to “geocentric” velocity at each of the tide gauge sites.
Importantly, all analysis results presented here include error margins at the 95% confidence limit (CL), unless noted otherwise. Key results are presented diagrammatically for each of the 58 stations analysed around the Australian coastline (see Figures 2–4). Analysis of the four (4) tide gauge records exceeding 75 years in length provided the means to examine the temporal characteristics in estimates of “relative” mean sea level and associated velocity and acceleration over time. From these records at the end of 2018 (see Table 2), the rate of “relative” sea-level rise varied between a low of 1.9 ± 1.6 mm/y (Fort Denison) and a maximum of 2.5 ± 1.1 mm/y (Port Adelaide, Outer). Of these records available for trend analysis, only Fort Denison had the highest mean sea-level velocity over the historical record available, occurring in 2018.
Figure 5 provides a detailed summary of the time-varying trend, velocity, and acceleration of the longest Australian records, Fort Denison and Fremantle, respectively, situated on either side of the country. Both exhibit similar temporal characteristics in which there are two distinctive peaks in the velocity charts (one in the first half of 20th century and one in the recent post-2000 portion of the record) and a distinct low point evident around 1980. Since the trough in the velocity record around 1980, both exhibit a steady increase in the “relative” velocity of ∼1.5 mm/y, suggesting a small acceleration in order to do so, though the measured accelerations are not statistically different to zero.
In general, results of the analysis of the other two records exceeding 75 years in length also exhibit similar temporal characteristics over the past 50–60 years, whereby a trough in the velocity time series occurs sometime in the period ∼1970–1990, and values increase thereafter to a peak occurring sometime after ca. 2010. In the case of Newcastle and Fort Denison over this period, the velocity increases continuously through to 2018.
The VLM analysis provides an important tool in estimating the influence of this factor on the rate of measured (or “relative”) sea-level rise at any specific tide gauge location. Negative VLMs indicate that the land margin is subsiding, therefore exacerbating the localised impacts associated with sea-level rise. Alternatively, positive VLMs indicate the land margin is rising, in turn moderating the influence of sea-level rise. Accordingly, locations with larger negative VLMs are at higher threat from the associated impacts of rising sea levels.
From the ALT-TG analysis undertaken here, large sectors of coastline around the Australian land mass are more prone to subsidence than uplift, heightening the general threat from sea-level rise into the future. At a broad scale, these include margins along the east coast between Townsville and Coffs Harbour, along the southern coast from Burnie to Port Pirie, and along the western and northwestern coastlines between Fremantle and Wyndham.
Of the 58 sites for which VLM estimates were determined, four (∼7%) were statistically greater than zero at the 95% CL (Cairns, Lord Howe Island, Eden, and Lorne), indicating the prevalence of uplift, whilst 23 sites (∼40%) were statistically lower than zero at the 95% CL, indicating the prevalence of subsidence. Of these sites, six showed a mean rate of subsidence exceeding 2 mm/y (Gold Coast, Hillarys, Bundaberg, Brisbane [West Inner Bar], Port Pirie, and Broome).
Of the seven sites for which comparative GNSS measured VLM values were available (see section on “Data Sources Used in This Study”), all agreed with the ALT-TG estimates at the 95% CL (Table 2; Figures 2–4).
With knowledge of VLMs, the “relative” velocity can be corrected to “geocentric” velocity for each of the long records (>75 years) which permit trend analysis (Table 2). A feature of interest is that the mean “geocentric” velocity in 2018 only exceeds 2 mm/y at Fort Denison (2.2 ± 1.9 mm/y).
Trends in SSH observed from multimission satellite altimetry data over the period from September 1992 to May 2019 (AVISO, 2019) show some key spatial signatures, both around the coastline and in the adjoining sea margins around the country. It should be clearly emphasised here that these are linear trends in SSH over the ∼26.5-year altimetry period, which will be significantly influenced by internal climate mode forcing mechanisms (such as El Niño–Southern Oscillation [ENSO], etc.) on such timescales. These trends are therefore not directly comparable to “relative” and “geocentric” velocities determined from the longer tide gauge record analysis in this study, which first removed such influences from the record and second estimated time-varying velocities in real-time, rather than averages across the record length.
Notwithstanding the above features, from the analysis of gridded products nearest the tide gauge sites no closer than 30 km from the open coast and around islands, the average trend in SSH across all 58 locations was approximately 3.3 mm/y. Some 64% of these locations exhibited a SSH trend exceeding 3 mm/y, located predominantly along the northern coastlines between Port Hedland and Bowen and along the length of the New South Wales (NSW) coast. Some 15 locations (26%) had SSH trends exceeding 3.5 mm/y over this period.
The altimetry SSH trend analysis highlighted large spatial variations in the sea margins around Australia. Figure 6 and Table 3 provide a summary of the SSH trends within the area bounded by longitudes 105°E and 160°E and latitudes 5°S and 50°S. Of the four quadrant areas (see Figure 6), the northern margins exhibited the highest average trends (>3.8 mm/y), with the southwest quadrant exhibiting the lowest average trend (∼2.7 mm/y). The highest trend of 11.1 mm/y was located in the southeast quadrant at 151.875°E and 38.875°S, representing the peak of a high trend mass situated in the Tasman Sea approximately 225 km southeast of the Victoria-NSW border. It is likely that this high trend mass is at least partially the response to a portion of the East Australian Current now extending further south, creating an area of more rapid warming in the Tasman Sea (BoM, 2018). Conversely, the lowest trend of –3.8 mm/y was located in the northwest quadrant at 131.875°E and 12.125°S, at the centre of a low trend margin situated in the Van Diemen Gulf region of the Timor Sea.
By employing state-of-the-art analytical techniques for mean sea-level research, the study gives rise to various discussion points, highlighted in the following sections.
Differentiating between Trends and Oscillatory Behaviour within Mean Sea-Level Records
As noted in the “Methods” section, understanding the transition point between the underlying trend (of primary interest) and low-frequency natural oscillations becomes the key consideration in analysing long mean sea-level time-series records.
There has been considerable evidence within the scientific literature for the presence of low-frequency bidecadal or multidecadal oscillations apparent within a variety of observational records and paleoclimate proxy data sets (e.g., Delworth and Mann, 2000; Folland, Palmer, and Parker, 1986; Folland, Parker, and Kates, 1984; Kerr, 2000; Knight et al., 2005; Knight, Folland, and Scaife, 2006; Kushnir, 1994; Mann and Emanuel, 2006).
The work of Schlesinger and Ramankutty (1994) and Mann and Park (1994, 1996) provided further evidence for a multidecadal (50–70 year) timescale signal centred in the North Atlantic with a weak projection onto hemispheric mean temperature. Based on analyses of paleoclimate proxy data, Mann, Park, and Bradley (1995) presented evidence that such a signal persists several centuries back in time (Mann, Steinman, and Miller, 2014).
With specific respect to sea levels, Chambers, Merrifield, and Nerem (2012) identified the existence of a significant oscillation with a period of around 60 years in the majority of the tide gauges examined during the 20th century, and they observed that it appears in every ocean basin, with amplitudes exceeding 20 mm in several long records. Chambers, Merrifield, and Nerem (2012) and Calafat and Chambers (2013) advised that estimates of global and regional acceleration must account for these multidecadal fluctuations.
The National Oceanic and Atmospheric Administration (NOAA) Atlantic Oceanographic and Meteorological Laboratory (AOML) (NOAA AOML, 2020) noted that with respect to the Atlantic Multidecadal Oscillation (AMO), studies of paleoclimate proxies, such as tree rings and ice cores, have shown that oscillations similar to those observed instrumentally have been occurring for at least the last millennium. This is clearly longer than modern humans have been affecting climate, so the AMO is probably a natural climate oscillation. In the 20th century, the climate swings of the AMO have alternately camouflaged and exaggerated the effects of global warming and made attribution of global warming more difficult to ascertain.
However, Mann, Steinman, and Miller (2020) observed a distinct (40–50 year timescale) spectral peak that appears in global surface temperature observations, which appears to reflect the response of the climate system to both anthropogenic and natural forcing rather than any intrinsic internal oscillation. Further, based on the available observational and modelling evidence, the most plausible explanation for the multidecadal peak seen in modern climate observations is that it reflects the response to a combination of natural and anthropogenic forcing during the historical era.
Mann, Steinman, and Miller (2020) advised that these findings have implications both for the validity of previous studies attributing certain long-term climate trends to internal low-frequency climate cycles and for the prospect of decadal climate predictability. In short, there is no evidence for a natural oscillation, either forced or unforced, that is present in the modern observational record. What has been inferred as an “AMO” signal in historical records is seen to be an artefact of analytical procedures that do not account for the structure in the total (both anthropogenic and natural) forced signal (Professor Michael Mann, personal communication, 14 April 2020).
On the basis of this new direction emerging in the scientific literature, it is prudent to consider that the difference between trend and oscillatory behaviour in mean sea-level data resides around the ∼50 year frequency band. Specifically, in the case of the analysis for this study, a trend has been considered to comprise components from an SSA decomposition of annual time series in which the substantive portion (>50%) of the relative spectral energy resides in the low-frequency band [0–0.2]. See the “Methods” section for more specific details.
Comparison with Previous Australian Sea-Level Studies
There have been numerous studies of Australian mean sea-level records undertaken to investigate trends, accelerations, and the spatial influence of climate modes and interdecadal variabilities (described in White et al., 2014). The well-cited study of White et al. (2014) provided a comprehensive understanding of mean sea-level changes and influencing factors around the Australian coastline based on data to the end of 2010. The study corrected for VLM using 12 GNSS records, which were similarly limited to the end of 2010, and extended the range of proximity between the GNSS and tide gauge by as much as 100 km. Key findings from this work included confirmation of considerable interannual and decadal variability in “relative” mean sea level at Australian gauges, with the largest values in the north and west (about 10–15 cm) and smallest values in the south and east. The variability was found to be coherent around the whole of the coastline and highly correlated with the Southern Oscillation Index. Higher sea-level trends observed across northern Australia were deemed to be largely associated with natural climate variability. The study observed that even after attempts to remove the effects of this natural variability, trends around most of Australia showed an increased rate of rise from the early 1990s, consistent with global mean trends.
An interesting element of the White et al. (2014) study was the fitted generalised additive model (GAM) assessment of the two longest records at Fort Denison (1886–2010) and Fremantle (1897–2010). The results showed the rate of rise to be nonlinear in nature, with both records indicating large rates of rise around the 1940s, comparatively stable relative mean sea levels between 1960 and 1990, and an increased rate of rise from the early 1990s.
Figure 7 highlights the comparison between the velocity outputs from the White et al. (2014) GAM assessment and the analysis undertaken in the current study. Although the timing of key changes in velocity aligns reasonably well, there are some considerable differences in the scale of the estimated velocities. One of the reasons for this is the differences in the underlying trend determination between the SSA technique and applied methodology (Watson, 2018) and those of the GAM technique. The latter technique is highly reliant on identifying, quantifying, and fitting key elements deemed to comprise the time series and more susceptible to less-well-modelled (or fitted) components transferring signal associated with more dynamic processes through to the resulting trend.
This present study highlights the comparatively low time-varying velocities and associated accelerations evident over the period of historical records around Australia. Until such time as the velocities and accelerations in mean sea level are sufficiently large not to be obscured by complex influences inducing decadal to multidecadal variability, complex VLM, and other background noise, the search for accelerations in ocean water-level records will require more intuitive, diagnostic considerations. For example, the measure of acceleration is perhaps more practically inferred by considering whether or not peaks in the instantaneous velocity and acceleration time series are increasing, are becoming more sustained, or are statistically abnormal (or different) over time in the context of the historical record. These diagnostic approaches will continue to be important until the extent of sea-level rise (due to climate change) is sufficient to be statistically differentiated from the remnant historical record with widespread spatial coherence (Watson, 2018). Whilst there is indeed spatial and temporal coherence in the likelihood of a weak acceleration in mean sea level evident around Australia after 1970–90, the evidence supporting the climate science and associated projection modelling from IPCC (2014) is not yet conclusive from the analysis and findings of this current study. This does not mean that Australian mean sea-level records are not exhibiting characteristics reflective of an already commenced climate change–induced acceleration, but such evidence must become more readily apparent from records around Australia over the next 20 years or so.
The current study augments the work of White et al. (2014) with some substantial improvements, including the addition of 8 years of additional tide gauge and satellite altimetry data, as well as a comprehensive national assessment of VLM using ∼26.5 years of ALT-TG–derived results. By comparison, the current study employed enhanced techniques to isolate the mean sea-level trend with improved temporal resolution, which have been specifically tailored and optimised for sea-level research (Watson, 2018).
Watson (2019a) observed that land motions embedded within tide gauge records are difficult contributions to resolve and isolate, in part because the general scale of VLM (see Ostanciaux et al., 2012) and the scale of sea-level rise trends due to climate change observed over the 20th and early 21st centuries are often similar (although the sign may differ). In attempting to convert “relative” to “geocentric” estimates of sea-level rise, the majority of contemporary studies make some allowance for land movements via the application of site-specific estimates of long-timescale GIA from the various models available (e.g., Lambeck, Smither, and Johnston, 1998; Peltier, 2004; Tushingham and Peltier, 1991).
However, GIA models provide only broad regional-scale estimates of the response of the earth's land masses to unloading associated with retreat of ice sheets since the Last Glacial Maximum around 26,500 years ago (Clark et al., 2009).
At more localised scales, numerous other processes contribute to VLM, including tectonics, volcanism, sediment compaction, and subsurface mineral, petrochemical, and water extraction, which are often of more significance than GIA (Zervas, Gill, and Sweet, 2013). Featherstone et al. (2015) determined VLM rates varying between –2 and –4 mm/y at Fremantle, Australia, resulting from subsidence due to increased groundwater abstraction from aquifers in the Perth Basin dating to the mid-1970s. Belperio (1993) determined significant, localised subsidence affecting the Port Adelaide estuary resulting from reclamation of Holocene wetlands and groundwater extraction from deeper Tertiary aquifers. In particular, the extraction of groundwater dating back as far as the early 1940s has created a potentiometric cone of depression in excess of 20 m below the estuary thought to contribute as much as 1.8–2.2 mm/y of subsidence, or up to three quarters of the secular sea-level rise measured at the Port Adelaide tide gauge (Belperio, 1993). The rates of VLM determined in the current study for the Perth Basin and Adelaide estuary accord well with estimates of subsidence from these earlier studies. Watson (2019b) noted that these rates of subsidence are an order of magnitude larger than rates of GIA calculated around the Australian mainland by White et al. (2014), which ranged from –0.2 to 0 mm/y.
In circumstances where the land is rising, the trend of “relative” sea level will be lowered; however, where subsidence is prevalent, the trend of “relative” sea level will be increased, exacerbating the local effects of a rising mean sea level. Land motions or neotectonic effects are known to be geographically highly variable (Harvey et al., 1999), and their quantification at tide gauge sites is an essential element in differentiating any secular (or greenhouse) signature from the tide gauge data. This is exemplified in the following discussion topic regarding the nonlinear VLM evident in the Newcastle record owing to mine subsidence (Watson, 2011; Watson et al., 2012).
This study presents the first national assessment of VLM using the ALT-TG technique, providing a scientifically robust estimate (Watson, 2019b) of land motion influence for sea-level studies, using a consistent approach based on ∼26.5 years of data records. Although coastal altimetry SSHs exhibit known limitations in proximity to the land-sea interface (Capet et al., 2014; Cazenave et al., 2017), continued advancements with the accuracy of SSH anomalies measured in coastal regions will only further advantage the current utility of ALT-TG proxies for VLM in coastal studies where GNSS stations are absent or lack comparatively lengthy or robust solutions.
Altimetry SSH Trends vs. Mean Sea-Level Rise from Long (Tide Gauge) Records
Multimission altimetry SSH trends are available from AVISO (2019) for the period spanning September 1992 to May 2019 (∼26.5 years). These trends were derived from comparatively simple linear regression analysis over the record length, and they are considered to incorporate all sea-level influences, including the strong interdecadal influences imposed by internal climate modes (such as ENSO, Pacific Decadal Oscillation [PDO], etc.), as well as rises in mean sea level due to longer-term influences, including those resulting from external climate forcing and anthropogenic greenhouse gas emissions.
White et al. (2014) concluded that higher sea-level trends observed across northern Australia were largely associated with natural climate variability. From analysis of tide gauge data and satellite altimetry, Gharineiat and Deng (2018) similarly observed that the interannual sea-level fingerprint in the northern Australian coastline is closely related to ENSO and Madden-Julian Oscillation events, with the greatest influence on the Gulf of Carpentaria, Arafura Sea, and Timor Sea. Such findings are clearly evident in the SSH trends around Australia presented in the “Results” section here.
Until the length of satellite altimetry data sets substantially increases, it is not yet possible to remove the highly contaminating influences associated with interannual to interdecadal processes from the records to estimate the underlying trend of primary interest. This trend is associated with mean sea level resulting from the influences of longer-term phenomena, which result in thermal expansion of the ocean water mass and melting of snow and ice reserves.
By comparison, mean sea level determined using the methodology applied to the long tide gauge records available for this study does permit the isolation and removal of all noise, high-frequency signals, and interannual and interdecadal signals (if the record is long enough). The resultant mean sea-level signal (or trend) is indeed representative of mean sea level resulting from the longer-term influences, which are of primary interest.
The midpoint of the linear trend analysis of SSH advised by AVISO (2019) occurs around 2006. In a comparison between this linear trend and the “geocentric” velocity at each of the long tide gauge records in 2006, the differences are quite significant (Table 4). Both sets of numbers are within a “geocentric” reference frame, but they are reflective of quite different record lengths and analysis techniques.
Subsidence in the Newcastle Record
Watson (2011) advised that one of the probable reasons for the clear disparity in the “relative” mean sea-level record between Newcastle and Fort Denison was the influence of underground mine subsidence in the vicinity of the Newcastle tide gauge. The current study, using improved time-series analysis techniques, provides greater insight into the nonlinear characteristics of the mine subsidence influence at Newcastle (Figure 8). From this analysis, the extent of the subsidence influence appears to have occurred largely prior to around 1969, resulting in this portion of the record increasing by ∼55 mm above that at Fort Denison. The maximum “relative” velocity in the Newcastle record of 5.2 ± 1.9 mm/y occurred in 1945, at a time when the “relative” velocity in the Fort Denison record was only 1.6 ± 0.4 mm/y. Both records exhibit similar temporal characteristics after ca. 1980 to present. GNSS receivers have been installed at both the Fort Denison and Newcastle tide gauges (since 2012 and 2013, respectively) to directly monitor precise land movements over time (Watson et al., 2012), but record lengths are currently relatively short for extrapolation of meaningful results.
How Do Rates of Mean Sea-Level Velocity around Australia Compare with Those Forecast from Current Climate Change Models?
Church et al. (2013) noted that the rate of rise at the start of the representative concentration pathway (RCP) experiments used for IPCC (2014) was about 3.7 mm/yr, which is slightly above the observational (altimetry) range of 3.2 ± 0.4 mm/y (90% CL) for 1993–2010, indicating that the simulated rate of climate warming is greater than that observed. For the RCP experiments reported in IPCC (2014), the start date correlating to the rate of rise of 3.7 mm/y was 2007.
Figure 9 (top panel) provides a summary of the “geocentric” velocities in 2007 at each of the four long record sites (exceeding 75 years) around Australia for comparison against the global estimate of 3.7 ± 0.4 mm/y (90% CL) used at the same time in IPCC (2014) for all the RCP projection model experiments to estimate forecasted sea-level rise. From this analysis, none of the stations accord with the global average estimate (95% CL). Perhaps more telling though, the average of the mean of the “geocentric” velocity of all the Australian station records in 2007 (1.1 mm/y) was ∼2.6 mm/y lower than the mean global average used in the RCP experiments for the same year (3.7 mm/y).
The bottom panel of Figure 9 provides a similar analysis but for the 2018 time frame against the current rate of global mean sea-level rise of 3.4 ± 0.4 mm/y (90% CL) (AVISO, 2019). From this analysis, three (of four) station records (Newcastle, Fort Denison, and Fremantle) accord with the global average estimate (95% CL). The increase in stations according with the global average is due in part to the wider error margins of the tide gauge analysis at the end of the record. However, the average of the mean of the “geocentric” velocity of the Australian station records (1.3 mm/y) remains nearly 2.1 mm/y lower than the current mean global average (3.4 mm/y). Noting that the global average advised from AVISO (2019) was determined as a linear regression over the time period from 1993 to 2019, if there were any acceleration in mean sea-level rise over this time frame, then the global rate in 2019 would be expected to be higher than the average linear rate over the time period, and therefore the actual difference (in 2019) would be expected to be even wider (than 2.1 mm/y).
Whilst these analyses are instructive, it should be noted that the time-series analysis applied to the available long tide gauge records removes high-frequency, interdecadal, and internal climate mode–driven signals first in order to estimate the residual trend representative of mean sea level resulting from the longer-term influences. By comparison, the global average rates of mean sea level advised from the linear regression over the altimetry record do not isolate and remove the strong contaminating influences associated with internal climate modes.
Is the Long Port Pirie Record Reliable for Long-Term Sea-Level Analysis?
The Port Pirie record began in 1941 and is one of the five longest Australian records in the public archives of the PSMSL available for trend analysis. At present, there are no identified quality-control issues associated with this particular record noted in the PSMSL station details. Figure 10 summarises the analysis of the Port Pirie (Spencer Gulf) and Port Adelaide (Gulf St. Vincent) records, which cover the same time frame (1941–2018) and are situated ∼180 km apart. This analysis readily highlights the unusual phenomena apparent in the Port Pirie record from about the mid-1990s onwards, showing a distinct upward change in the mean sea-level trend that drives the “relative” velocity from zero to a maximum rate of 7.3 ± 2.9 mm/y in only 20 years. Whilst key interannual signatures are temporally aligned between the two records (as expected from their regional proximity), the trajectory of the mean sea-level trends varies significantly. Separate analysis undertaken by the Tidal Unit of the Bureau of Meteorology, which compared monthly differences between Port Pirie and five other regional tide gauge records, suggested a step change evident in the Port Pirie record relative to the other sites around 1999 (James Chittleborough, Assistant Manager, Tidal Unit, Bureau of Meteorology, personal communication). Until such time as the underlying reason for the apparent sharp change in the Port Pirie record during the mid-late 1990s is better understood and adequately accounted for, evidence suggests that the utility of this record for long-term sea-level analysis should be viewed with considerable caution.
The potential threats to Australia from current and projected sea-level rise are significant and have profound environmental, social, and economic consequences. The current threats associated with coastal hazards (e.g., coastal erosion, storm surge, oceanic inundation, etc.) will be exacerbated by projected sea-level rise associated with climate change, which is anticipated to increase at an increasing rate over the course of the 21st century (and beyond).
IPCC (2014) mean sea-level rise projections (Church et al., 2013) based on modelling of various RCPs (Moss et al., 2010; Van Vuuren et al., 2012) range from ∼20 to 100 cm over the course of the 21st century (relative to 1990). The possibility of larger projections has also been observed in the literature (e.g., Hansen et al., 2016; Kopp et al., 2017). A key issue is the fact that the influence on projected sea levels of the radiative forcing built into each RCP experiment only really starts to diverge significantly beyond the late 2030s during the 21st century.
Thus, temporal characteristics of the projected rate of global mean sea-level rise provide only a coarse reference frame against which long tide gauge and other records (such as satellite altimetry) can be compared to augment scientific understanding and adaptive planning endeavours at a local or regional scale over the next two decades (in particular). Given the threat posed by this phenomenon, every effort must be made to routinely monitor and review sea-level data around the country (and world), enabling the early detection of key emerging trends of significance that will aid coastal planning, design, and risk management activities.
This current study was augmented by the first comprehensive national assessment of VLMs, which provide key insights into margins around the coastline where subsidence is prevalent, exacerbating the localised effects of a rising global mean sea level. The analysis revealed that at a broad scale, margins along the east coast between Townsville and Coffs Harbour, along the southern coast from Burnie to Port Pirie, and along the western and northwestern coastlines between Fremantle and Wyndham are more subject to subsidence.
This study updated and extended the several previous works undertaken to analyse tide gauge records and satellite altimetry around Australia using enhanced time-series analysis techniques to isolate mean sea level with greater precision, in turn providing improved estimates of the rate of associated time-varying velocities and accelerations to present. Four records exceeding 75 years in length, meeting strict quality-control requirements (see “Methods”), were analysed to chart time-varying “relative” velocity and accelerations up to and including 2018 at each location. Averaged across these four records in 2018, approximately 40% of the “relative” velocity observed (∼2.2 ± 1.8 mm/y, 95% CL) is attributable to VLM. When corrected for VLM, only the Fort Denison site exhibits “geocentric” mean sea-level velocity in 2018 exceeding 2 mm/y. The average “geocentric” velocity across all four sites in 2018 equates to 1.3 ± 2.0 mm/y (95% CL).
A feature of interest is that all four sites where record lengths exceed 75 years generally exhibit similar temporal characteristics over the past 50–60 years, whereby a trough in the velocity time series occurs sometime in the period 1970–1990, and values increase thereafter to a peak occurring sometime after ca. 2010, suggesting the presence of a small acceleration (albeit not statistically different to zero at the 95% CL) in the record. However, only the Fort Denison record exhibits the highest velocity observed over the full length of the record in 2018.
From this long record analysis in 2018, three (of four) station records accord with the global average estimate (95% CL); however, the mean “geocentric” velocity of the Australian station records (1.3 mm/y) remains some 2.1 mm/y lower than the global average rate of rise of 3.4 mm/y based on satellite altimetry (AVISO, 2019).
The study of mean sea levels and associated climate change–induced projection modelling are indeed complex elements of science. Of the information resources that we have at our disposal, we know that the coastal margins more vulnerable to the threats posed by rising sea levels are those in which subsidence is prevalent, higher satellite altimetry SSH trends are evident, and higher “geocentric” velocities are being observed. The evidence from this study suggests that the more heavily populated eastern coastline of Australia is exposed to more of these factors than other coastlines at present.
Currently, the areas of the Tasman Sea exhibiting elevated SSH trends (maximum of ∼11 mm/y) during the altimetry era are not yet manifesting themselves in associated high increases in “geocentric” velocity from the long tide gauge record analysis at Fort Denison or Newcastle. However, should this area of more rapid warming in the Tasman Sea (BoM, 2018) continue to expand and encroach closer to the coastline, as anticipated under climate change predictions (Zhang et al., 2017), these impacts will be readily detected in the records at these stations using the analysis techniques advised in this study.
With continually improving and lengthening satellite altimetry and tide gauge records, the projected climate change influence on sea levels around Australia will become more readily apparent over the next two decades. An update of the analysis and results presented herein is recommended in 5 years' time to identify and monitor key changes.
The author would like to thank the following: Mike Davis (data controller, Tidal Unit, Australian Bureau of Meteorology), for facilitating the provision of Australian tide gauge station data up to December 2018; James Chittleborough (assistant manager, Tidal Unit, Bureau of Meteorology), for analysis and discussions regarding the Port Pirie record; Emeritus Professor Nick Harvey (University of Adelaide), for discussions regarding the Port Pirie record; Professor Michael Mann (director, Earth System Science Center, Penn State University, University Park, Pennsylvania), for discussions concerning separating natural oscillations from underlying trends; Doug Lord (principal and director, Coastal Environment Pty., Ltd.), for review and comments that improved the paper; Honorary Associate Professor Ron Cox (School of Civil and Environmental Engineering, University of NSW, Kingsford, NSW, Australia), for review and comments that improved the paper; the European Commission Copernicus Marine Environment Monitoring Service (CMEMS), for access to altimetry products and services used in this study; and the Permanent Service for Mean Sea Level, for their publicly accessible data repository.