Andrew C. Kemp’s research while affiliated with Tufts University and other places

We propose a Bayesian, noisy-input, spatial–temporal generalized additive model to examine regional relative sea-level (RSL) changes over time. The model provides probabilistic estimates of component drivers of regional RSL change via the combination of a univariate spline capturing a common regional signal over time, random slopes and intercepts capturing site-specific (local), long-term linear trends and a spatial–temporal spline capturing residual, non-linear, local variations. Proxy and instrumental records of RSL and corresponding measurement errors inform the model and a noisy-input method accounts for proxy temporal uncertainties. Results highlight the decomposition of regional RSL changes over 3,000 years along North America’s Atlantic coast. The physical process glacial isostatic adjustment prevailed before 1800 CE, with anthropogenic forcing dominating after 1900 CE.

Bulk organic stable carbon isotope (δ 13 C) and accompanying bulk organic geochemical measurements have been increasingly used as a relative sea-level (RSL) indicator over the last two decades. Their utility as a RSL indicator is premised on the ability of bulk organic δ 13 C and bulk organic geochemistry to distinguish between organic matter (OM) sources in coastal environments, and to identify changes in OM source contributions in sediment sequences in response to RSL change. We evaluate the performance of bulk δ 13 C and bulk organic geochemistry as a RSL indicator in tidal marsh, mangrove and isolation basin environments. The interpretation of isotope measurements from these environments requires knowledge of the processes controlling contemporary OM δ 13 C, and the influence of decomposition on bulk values. We review in detail the controls on the δ 13 C composition of OM in tidal marshes, mangroves and isolation basins, and advocate wherever possible for the collection of contemporary geochemical datasets corrected for the 13 C Suess effect to help inform interpretations. From the wide range of case studies considered, an emerging principle is that the degree of isotopic distinctiveness between OM sources is key in determining how the technique can be deployed as a RSL indicator. This can range in use from the provision of qualitative information on changes in marine influence over time, to the identification of sea-level index points at lithostratigraphic contacts, and most powerfully to the recognition of inter-tidal sub-environments with isotopically well-constrained vertical limits.

We propose a Bayesian, noisy-input, spatial-temporal generalised additive model to examine regional relative sea-level (RSL) changes over time. The model provides probabilistic estimates of component drivers of regional RSL change via the combination of a univariate spline capturing a common regional signal over time, random slopes and intercepts capturing site-specific (local), long-term linear trends and a spatial-temporal spline capturing residual, non-linear, local variations. Proxy and instrumental records of RSL and corresponding measurement errors inform the model and a noisy-input method accounts for proxy temporal uncertainties. Results focus on the decomposition of RSL over the past 3000 years along the Atlantic coast of North America.

Tide-gauge measurements in the western North Atlantic Ocean show coherent, multi-decadal relative sea-level (RSL) trends across multiple spatial scales. Proxy reconstructions developed from salt-marsh sediment can extend this instrumental record. However, the degree of coherence in proxy reconstructions is underexamined through within-region replication. To explore within-region replication, we developed a new RSL reconstruction from Fox Hill Marsh, Rhode Island to complement similar records at nearby sites. We established the elevation of former sea level from assemblages of foraminifera and bulk-sediment δ¹³C values using a Bayesian transfer function. We employed radiocarbon dating and recognition of pollution horizons to construct a core chronology. Since ∼1200 BCE, RSL rose by ∼3.7 m at Fox Hill Marsh. After correction for glacial isostatic adjustment, application of a statistical model intended to quantify (multi-) century-scale trends showed that the fastest rate of rise in at least the past 3000 years was 1.71 ± 0.84 mm/yr (95% credible interval) in 2020 CE. This result replicates regional tide-gauge measurements and other proxy reconstructions. Using an alternative statistical model constructed to identify sub-centennial sea-level changes, we examined if there was a hotspot of 18th century rise in the northeastern United States and found no spatially-coherent trend (i.e., occurring at all or most sites). This lack of replication indicates that accelerated rise during the 18th century is likely local (site-specific) in scale, or an artifact of individual reconstructions. Continued efforts to replicate RSL reconstructions will increase confidence in the accuracy of records and their subsequent interpretation.

Beginning ~3,500 to 3,300 y B.P., humans voyaged into Remote Oceania. Radiocarbon-dated archaeological evidence coupled with cultural, linguistic, and genetic traits indicates two primary migration routes: a Southern Hemisphere and a Northern Hemisphere route. These routes are separated by low-lying, equatorial atolls that were settled during secondary migrations ~1,000 y later after their exposure by relative sea-level fall from a mid-Holocene highstand. High volcanic islands in the Federated States of Micronesia (Pohnpei and Kosrae) also lie between the migration routes and settlement is thought to have occurred during the secondary migrations despite having been above sea level during the initial settlement of Remote Oceania. We reconstruct relative sea level on Pohnpei and Kosrae using radiocarbon-dated mangrove sediment and show that, rather than falling, there was a ~4.3-m rise over the past ~5,700 y. This rise, likely driven by subsidence, implies that evidence for early settlement could lie undiscovered below present sea level. The potential for earlier settlement invites reinterpretation of migration pathways into Remote Oceania and monument building. The UNESCO World Heritage sites of Nan Madol (Pohnpei) and Leluh (Kosrae) were constructed when relative sea level was ~0.94 m (~770 to 750 y B.P.) and ~0.77 m (~640 to 560 y B.P.) lower than present, respectively. Therefore, it is unlikely that they were originally constructed as islets separated by canals filled with ocean water, which is their prevailing interpretation. Due to subsidence, we propose that these islands and monuments are more vulnerable to future relative sea-level rise than previously identified.

Late Holocene relative sea-level reconstructions are commonly generated using proxies preserved in salt-marsh and mangrove sediment. These depositional environments provide abundant material for radiocarbon dating in the form of identifiable macrofossils (salt marshes) and bulk organic sediment (mangroves). We explore if single-step graphitization of these samples in preparation for radiocarbon dating can increase the number and temporal resolution of relative sea-level reconstructions without a corresponding increase in cost. Dating of salt-marsh macrofossils from the northeastern United States and bulk mangrove sediment from the Federated States of Micronesia indicates that single-step graphitization generates radiocarbon ages that are indistinguishable from replicates prepared using traditional graphitization, but with a modest increase in error (mean/maximum of 6.25/15 additional ¹⁴ C yr for salt-marsh macrofossils). Low ¹² C currents measured on bulk mangrove sediment following single-step graphitization likely render them unreliable despite their apparent accuracy. Simulated chronologies for six salt-marsh cores indicate that having twice as many radiocarbon dates (since single-step graphitization costs ∼50% of traditional graphitization) results in narrower confidence intervals for sample age estimated by age-depth models when the additional error from the single-step method is less than ∼50 ¹⁴ C yr (∼30 ¹⁴ C yr if the chronology also utilizes historical age markers). Since these thresholds are greater than our empirical estimates of the additional error, we conclude that adopting single-step graphitization for radiocarbon measurements on plant macrofossils is likely to increase precision of age-depth models by more than 20/10% (without/with historical age markers). This improvement can be implemented without additional cost.

Salt-marsh sediment is an important geological archive for reconstructing Holocene relative sea level. The vertical uncertainty of these reconstructions is usually quantified as proportional to tidal range, resulting in a presumed primacy of microtidal regions for generating precise records. We analyzed hourly water-level predictions and measurements over 1983–2001 from 43 tide gauges on the U.S. Atlantic coast to quantify the (1) relationship between elevation (relative to tidal datums) and duration of inundation; and (2) relative importance of astronomical tides and non-tidal (meteorological, hydrologic, and oceanographic) processes. Non-tidal processes are more likely to exert a proportionally large influence on water levels in regions with small astronomical tides, and less likely to do so where tides are large. In some microtidal regions (e.g., Chesapeake Bay), the precision of RSL reconstructions is likely overestimated, while in regions with large tidal range (e.g., Gulf of Maine) it may be underestimated. Adopting an inundation frequency rather than predicted astronomical datum for the upper limit of salt-marsh distribution may reduce this bias. Relative sea-level reconstructions often assume constant tidal range through time, but in some microtidal regions this assumption should be expanded to explicitly describe additional stationarity of non-tidal contributions.

A paucity of detailed relative sea-level (RSL) reconstructions from low latitudes hinders efforts to understand the global, regional, and local processes that cause RSL change. We reconstruct RSL change during the past ~5 ka using cores of mangrove peat at two sites (Snipe Key and Swan Key) in the Florida Keys. Remote sensing and field surveys established the relationship between peat-forming mangroves and tidal elevation in South Florida. Core chronologies are developed from age-depth models applied to 72 radiocarbon dates (39 mangrove wood macrofossils and 33 fine-fraction bulk peat). RSL rose 3.7 m at Snipe Key and 5.0 m at Swan Key in the past 5 ka, with both sites recording the fastest century-scale rate of RSL rise since ~1900 CE (~2.1 mm/a). We demonstrate that it is feasible to produce near-continuous reconstructions of RSL from mangrove peat in regions with a microtidal regime and accommodation space created by millennial-scale RSL rise. Decomposition of RSL trends from a network of reconstructions across South Florida using a spatio-temporal model suggests that Snipe Key was representative of regional RSL trends, but Swan Key was influenced by an additional local-scale process acting over at least the past five millennia. Geotechnical analysis of modern and buried mangrove peat indicates that sediment compaction is not the local-scale process responsible for the exaggerated RSL rise at Swan Key. The substantial difference in RSL between two nearby sites highlights the critical need for within-region replication of RSL reconstructions to avoid misattribution of sea-level trends, which could also have implications for geophysical modeling studies using RSL data for model tuning and validation.