Snapshots of Coastal Ecology Using Multiproxy Analysis Reveals Insights Into the Preservation of Swamp and Marsh Environments Since the Late Pleistocene

Abstract

The southeastern United States Coastal Plain ecosystem contains baldcypress (Taxodium distichum) swamps and grass-dominated marshes. These ecosystems also occurred on the exposed continental shelf during lower sea levels but are rarely preserved due to the mechanically erosive nature of transgression and regression. Two presently marine sites on the northeastern Gulf of Mexico's continental shelf contain well-preserved woody terrestrial sediments that were the subject of previous studies. This study continues the investigation using geochemical (δ¹³C, δ¹⁵N, δ³⁴S) and palynological characteristics of these formerly terrestrial sediments to determine if swamps and/or marshes existed at the time of deposition. The first site is located ∼20 km southeast of Horn Island, Mississippi (MS) and the core has terrestrial sediments radiocarbon dated to 11,066–10,228 (2σ) calibrated years BP (early Holocene). The second site is the “Alabama Underwater Forest” located ∼13 km south of Gulf Shores, Alabama (AL) and the cores have terrestrial sediments optically stimulated luminescence dated to 63 ka (±10 ka, 2σ) to 72 ka (±16 ka, 2σ) (late Pleistocene). Geochemical results for the MS sediments indicate a swamp-to-freshwater marsh transitional series, whereas the AL sediments indicate a swamp-to-saltwater marsh transitional series, both supported by palynological results. Further exploration of the geochemical results using linear discriminant analysis, trained with published geochemical data, supports the swamp and marsh interpretations. We conclude that the near-pristine preservation of these woody deposits is not solely due to physical mechanisms, such as rapid burial, but is also coupled with anoxia- and euxinia-driven biogeochemical reactions promoting wood and woody debris preservation in swamp and marsh environments.

Key Points

• Multiple paleo-terrestrial deposits containing wood and peat are preserved offshore in the northeastern Gulf of Mexico

• Geochemical and microfossil analyses find that these deposits are past swamp and marsh ecosystems

• Swamps and marshes that are anoxic and/or euxinic promote the biogeochemical preservation of woody and organic matter

Plain Language Summary

Scientists have studied unique sites offshore of Alabama and Mississippi that contain soils, peat, wood, and tree stumps. Scientists have determined that the trees from the Alabama site are between 42,000 and 72,000 years old and were alive during the last ice age when sea levels were much lower than today. The Mississippi site contains woody peat sediments about 11,000 years old and occurred when sea levels were rapidly rising globally. These types of formally land-based sediments are eroded as the sea level rises at the end of the glacial periods. Furthermore, wood and plant matter decay after the plant dies and become part of the soil. We investigated the chemistry of the preserved soils to understand the conditions that lead to their preservation. Chemical and pollen analyses found that these deposits are past swamp and marsh ecosystems that occurred on land similar to those occurring along the United States Gulf Coast today. Swamps and marshes have still waters with little oxygen that stop decay leading to the preservation of the wood and tree stumps. Swamps occur in depressions where water can collect and over time they fill with sediments. These preserved swamp deposits are time capsules from the past.

1 Introduction

The Gulf Coastal Plain of the southeastern United States is home to a complex system of low-elevation environments, such as baldcypress (Taxodium distichum) swamps and marshes, all interacting simultaneously with continental and marine geomorphic environments, such as barrier islands, deltas, estuaries, and river systems (Battaglia et al., 2012). During glacial advances throughout the Pleistocene (Figure 1a), shorelines advanced seaward and exposed new land on previously submerged continental shelves (Figure 1b) (Clark & Mix, 2002; Peltier, 1994), and their associated environments shifted with the coastlines (Bartek et al., 2004; Flocks et al., 2011; Hollis et al., 2019). However, the depositional history of these terrestrial sediments is rarely preserved in the strata due to the mechanically erosive nature of transgressions and regressions in response to sea level fluctuations (Cahoon et al., 2006; Swift, 1968). To gain insight into these past coastal environments, study sites would also need to contain undisturbed and near-pristine preservation of past terrestrial materials, meaning that micro- and macrofossils were not degraded and altered via physical, chemical, or mineralogical processes to the point of disrupting ecological signals, which are not typically found along the southern Coastal Plain (Bryant & Holloway, 1985; Delcourt & Delcourt, 1987).

Figure 1

Locations of study sites in the Gulf of Mexico of the southeastern United States (inset map). (a) Global sea level relative to modern for the past 80 ka (black line) with estimated maximum and minimum sea level (shaded gray area) (Waelbroeck et al., 2002). Marine Isotope Stages 1–5a (gray dashed lines) are based on Railsback et al. (2015). Age ranges for the Mississippi (MS; purple area) and Alabama (AL; yellow area) sites are noted on the age axis and the tops of the terrestrial sediment depth as horizontal lines. Gray circles are ¹⁴C dates (2σ) and black squares are optically stimulated luminescence (OSL) dates (1σ), respectively, shown with respect to dating sample depth (Figures S1–S3 in Supporting Information S1) (DeLong et al., 2021; Miller et al., 2021). For the AL site, the age range shown is for OSL dates from the bottom of the cores with Pleistocene terrestrial sediments and ¹⁴C dates in the transitional Holocene facies or top of the Pleistocene terrestrial sediments; however, seven of the nine sediment samples are radiocarbon dated as >43.5 ka BP, or radiocarbon dead (Figures S2 and S3 in Supporting Information S1) (DeLong et al., 2021). (b) Estimated coastlines (blue lines) for past sea levels assuming modern bathymetry. Topography is indicated via shading and isobaths by gray lines at 10-m increments. The MS (purple) and AL (yellow) sites are labeled on the −10 m map. White lines indicate modern state lines (Florida is FL) and modern coastlines. Exact coordinates are not provided to protect the sites.

Two marine sites on the northeastern Gulf of Mexico's outer continental shelf (Figure 1) have been identified that contain terrestrial sediments containing well-preserved pollen, woody peat deposits, and in one site, tree stumps in life position exposed on the seafloor (see Section 1.1) (DeLong et al., 2021; Miller et al., 2021; Reese et al., 2018a). These sites were previously inland terrestrial environments on the exposed continental shelf (i.e., land) when the sea level was lower (Figure 1). Their preservation is due in part to favorable geomorphology at the time of deposition, that is, depressions with ample accommodation space, ample sedimentation rates, and/or rapid burial (DeLong et al., 2021; Ethridge et al., 1987). Furthermore, biogeochemical processes that promote preservation (e.g., anoxia), which are innate to the Gulf Coastal Plain environments such as swamps and marshes, have been proposed as an additional contributing factor in the preservation of organics and woody deposits in these unique terrestrial sediments now located in the marine environment (DeLong et al., 2021). Despite local physical mechanisms that promote the preservation of these terrestrial sediments and organic matter, the same level of preservation is seldom found in marine and coastal areas (e.g., Lee et al., 2019; Pearson et al., 1986) due to erosive processes and decomposition of organic matter.

These sedimentary archives preserved on the Gulf of Mexico's continental shelf could provide insight into past environmental settings of the region (DeLong et al., 2021; Miller et al., 2021). Stable isotope ratios (e.g., δ¹³C, δ¹⁵N, and δ³⁴S) across environments provide insight into elemental cycling and allow for the investigation of past ecological conditions (Fry, 2006). Any single biogeochemical indicator, such as δ¹³C, provides environmental information (e.g., C₃ vs. C₄ plants), yet a multiproxy approach with several indicators coupled with pollen analysis can improve interpretations. This study analyzed δ¹³C, δ¹⁵N, and δ³⁴S profiles in the terrestrial sediments at these two sites along with the pollen results to understand ecosystem shifts with time and paleoenvironmental conditions. The δ¹³C reveals shifts in C₃ (most trees and plants) to C₄ (drought-tolerant plants like corn and grasses) in the environment. Thus, as the sites transition from a wetland forest (C₃) to saltwater grasses (C₄), there will be a shift in δ¹³C. The δ¹⁵N provides information on nutrients, specifically for this research, the presence or absence of freshwater or marine algae. The δ³⁴S discerns changes from terrestrial to marine conditions. This study seeks to elucidate the biogeochemical properties of these formally terrestrial environments that contributed to their excellent preservation of pollen, peat, woody debris, and tree stumps.

1.1 Background and Motivation

In the fall of 2004, a site located ∼13 km south of Gulf Shores, AL (Figure 1b) was found to contain many tree stumps rooted in situ that were well-preserved with no evidence of petrification and were found to be radiocarbon dead or older than 43.5 ka BP (DeLong et al., 2021; Raines, 2017). Hurricane Ivan, which made landfall on the AL coast on 16 September 2004, generated enough wind and wave activity to scour the seafloor (Teague et al., 2006; D. W. Wang et al., 2005) and thus is inferred as the mechanism that exposed the Pleistocene-aged forest previously buried in the subsurface. In the years following the discovery of this site, it has become colloquially known as the Alabama Underwater Forest (Raines, 2017) and is the subject of ongoing investigations since 2012 (DeLong et al., 2020, 2021; Garretson, 2022; Gonzalez, 2018; Gonzalez et al., 2017; Moran, 2023; Moran et al., 2024; Obelcz, 2017; Obelcz et al., 2023; Reese et al., 2018a; Truong, 2018). The MS study area where woody peat was found in sediment cores collected in 2019 is still buried and not exposed on the seafloor to our knowledge (Miller et al., 2021). That site is similar to the AL site in depth and sediment types; therefore, this study applies similar analyses to the MS site for comparison.

Palynological analyses of the Pleistocene-aged facies in the AL cores (15DF-1 and 15DF-3B), which contained higher organic material (Figures S2 and S3 in Supporting Information S1) (DeLong et al., 2021; Gonzalez et al., 2017), were identified as a Cypress/Tupelo forest and Cypress/Alnus swamp communities bracketed by open marsh with grasses that could be freshwater or saltwater, similar to present-day Gulf Coastal Plain environments (Figures S5 and S6 in Supporting Information S1) (Garretson, 2022; Reese et al., 2018a). As the sea level rises, coastal plant communities shift from freshwater- to saltwater-grass dominated, so discerning the type of marsh grasses aid in interpretations related to sea level changes. Unfortunately, palynological analyses cannot discern between freshwater or saltwater grasses; therefore, this study uses stable isotope ratios (δ¹³C, δ¹⁵N, and δ³⁴S) coupled with palynological and sedimentology analyses to discern the paleoenvironmental setting at these sites, namely freshwater versus saltwater marshes, associated with changes in sea level. Additionally, this study conducts palynological analysis at the MS site to provide information on the plant ecosystems presented at the time of deposition. Such analyses can help to refine and strengthen the palynological and paleoenvironmental interpretations, as observed in the previous microfossil assemblage studies (DeLong et al., 2021; Garretson, 2022; Reese et al., 2018a) and are necessary for understanding the possible chemical mechanisms that may have contributed to terrestrial sediment and woody debris preservation.

The goals of this study are to reconstruct environmental settings for terrestrial deposits (i.e., forested swamp, freshwater, or saltwater grass marshes) in the northeastern Gulf of Mexico using multiproxy approaches, and to elucidate biogeochemical properties of these environments to provide insight into the overall reason for the preservation of the terrestrial organic matter.

1.2 Study Areas

The northeastern Gulf of Mexico along the Mississippi and Alabama coasts is characterized as a slowly subsiding, passive continental margin (Sydow & Roberts, 1994). Glacio-isostatic uplift and subsidence rates are not known for the sites presented in this study, but are likely to be on the order of ±2 m during the last glacial interval (Blum et al., 2008). Region-specific sea level curves for the northern Gulf of Mexico are not well constrained beyond 10 ka (e.g., Milliken et al., 2008), and thus many studies have sought to contribute to refining the region's sea level estimates by accounting for local factors as they relate to preserved sedimentary environments, glacial isostatic adjustments, etc. (e.g., Anderson et al., 2014, 2016; Simms et al., 2007). Therefore, we can assume modem bathymetry as a good estimate of past coastlines using global sea level estimates (Figure 1b) (Waelbroeck et al., 2002).

Two sites in the northeastern Gulf of Mexico (Figure 1b) have been identified to contain muddy-peat terrestrial sediments with woody debris that are well-preserved (DeLong et al., 2021; Miller et al., 2021). The western site is ∼20 kilometers (km) southeast of Horn Island, MS, at 21.1 meters (m) below sea level (mbsl) and the eastern site is ∼13 km south of Gulf Shores, AL, at ∼15 mbsl. Both sites (∼85 km apart) are located on the outer continental shelf between and above the Mississippi River mouth and north of DeSoto Canyon within the Mississippi-Alabama-Florida (MAFLA) sand sheet province (Doyle & Sparks, 1980; McBride et al., 1999) deposited during the Holocene from sandy sediments discharged by small rivers in the area (Anderson et al., 2004). Below the MAFLA sand sheet are lowstand deeply incised valleys carved out by the Mobile-Tensaw River and Biloxi-Pascagoula River systems during the Last Glacial Maximum (Marine Isotope Stages (MIS) 2) lowstand (Bartek et al., 2004; Gal et al., 2021; Hollis et al., 2019). The AL site is located to the north of the eastern incised valley of the Mobile-Tensaw river system (Bartek et al., 2004; DeLong et al., 2021; Obelcz, 2017) and the MS site is located in the Biloxi-Pascagoula River incised valley system (Miller et al., 2021). The incised paleovalleys extending seaward contribute to the geomorphology of the barrier island system in this region (Bartek et al., 2004; Flocks et al., 2011; Gal et al., 2021) as well as the geomorphic responses to sea level rise and associated transgressive ravinement (Hollis et al., 2019).

One sediment core (19OCS-SI39) was selected from the MS site and two cores (15DF-1 and 15DF-3B) from the AL site to be analyzed for this study (Figure 1b; Table S1 in Supporting Information S1); we do not provide exact coordinates to protect the sites. The core's sedimentological analyses and dating are described in previous studies (Figures S1–S4 in Supporting Information S1) (DeLong et al., 2021; Gonzalez et al., 2017; Miller et al., 2021). The cores for this study were selected based on defined lithologic layers (e.g., no visible bioturbation or mixing) and because they have terrestrial sediments containing woody debris and peat (DeLong et al., 2021; Miller et al., 2021). This study focuses on the downcore sections containing terrestrial sediments with woody debris (sandy mud and interbedded mud-peat facies) and excludes marine depositional facies overlaying an erosional surface. The terrestrial sediments in the MS core are dated to 11,066 to 10,228 (2σ) calibrated (cal) year BP starting at 25.14 mbsl (Miller et al., 2021). For the AL site, the terrestrial sediments were optically stimulated luminescence (OSL) dated to ∼72 ± 8 ka (1σ) to 56 ± 5 ka (1σ) across the site (DeLong et al., 2021). Core 15DF-3B had one sample at 18.9 mbsl radiocarbon dated, it was >43.5 ka BP or radiocarbon dead (DeLong et al., 2021). Core 15DF-1 had 10 samples radiocarbon dated, the sample from the marine sand section was 3,920 cal BP confirming the Holocene sand facies, and only two samples from the interbedded muddy-peat facies returned radiocarbon ages (45.2 and 41.8 cal ka BP), the others were radiocarbon dead, thus confirming the interbedded muddy-peat facies is late Pleistocene (Figure 1a) (DeLong et al., 2021; Gonzalez et al., 2017; Reese et al., 2018a. Using global sea level estimates (Figure 1a) (Waelbroeck et al., 2002), the MS site has an estimated sea level at the time of terrestrial sediment deposition ranging from −45.0 to −34.7 m (−58 to −21.7 m with 2σ uncertainties). A regional relative sea level curve using glacio-hydro-isostatic models and observational data suggests an estimated sea level of approximately −30 m between ∼10 and 11 ka for nearby Mobile Bay, AL, although these estimates extend back only to 20 ka and have considerable uncertainty (Simms et al., 2007). The AL site has an estimated sea level at the time of terrestrial sediment deposition ranging from −85 to −48 m (−98 to −7 m with 2σ uncertainties). Despite these large uncertainties in dating and sea level estimates, both sites were subaerial terrestrial environments on the exposed continental shelf when the sea level was lower.

2 Methods

Core 19OCS-SI39 was collected as part of research efforts to expand the mapping of the MIS 2 surface offshore of MS (Miller et al., 2021). The geophysical surveys and coring trip were conducted in 2019 directly south of the original study area as reported by Gal et al. (2021) and Hollis et al. (2019). Multiple sediment cores were recovered during the 2019 vibracoring field expeditions onboard the R/V Apalachee (Florida State University Coastal & Marine Laboratory) (Culver-Miller et al., 2021; Miller et al., 2021) that contain well-preserved terrestrial woody peat and wood deposits. Core 19OCS-SI39 contains interbedded muds and woody peat at ∼4 m depth in the core or 24.8 to 25.6 mbsl (Figure S1 in Supporting Information S1). The marine sand facies in core 19OCS-SI39 overlying floodplain deposits and interbedded mud and peat is indicative of an erosional surface (Miller et al., 2021). The interpretation of terrestrial deposits is based on geophysics, sedimentological analyses, lack of shell and sand material, and correlation of similar age deposits across four cores (Miller et al., 2021). The core is stored at the University of Southern Mississippi's School of Ocean Science and Engineering.

The cores 15DF-1 and 15DF-3B (Figures S2–S4 in Supporting Information S1) were collected in 2015 from the AL site using a 5-m vibracorer and tripod deployed from the R/V Coastal Profiler of the Coastal Studies Institute at Louisiana State University (LSU) (DeLong et al., 2020, 2021). Additionally, geophysical surveys were conducted in 2015, 2016, and 2021 to collect side-scan sonar and CHIRP sub-bottom profiles as well as bathymetric data at the AL site (DeLong et al., 2021; Moran, 2023; Obelcz, 2017). Geophysical, lithological, and microfossil analyses found the cores contained ∼1 m of accumulated terrestrial deposits of fine-grained material (sand and mud) and organic matter (muddy peat) accumulating in depressions of a swamp environment and/or fluvial floodplain where sediments accumulated (DeLong et al., 2021; Obelcz, 2017). These sediment cores have an unconformity above the Pleistocene-aged deposits that are covered by a ∼3-m layer of Holocene-aged sand (DeLong et al., 2021; Gonzalez et al., 2017). This study focuses on the downcore sections containing Pleistocene-aged terrestrial sediments and excludes marine depositional facies (DeLong et al., 2021; Gonzalez et al., 2017; Miller et al., 2021). The AL cores are archived in cold storage at the LSU Department of Geology and Geophysics.

2.1 Sample Preparation

Cores 19OCS-SI39, 15DF-1, and 15DF-3B were sampled for ∼10 g of sediment at a 5-cm interval in the terrestrial facies based on lithological core descriptions (DeLong et al., 2021; Gonzalez et al., 2017; Miller et al., 2021) (Figures S1–S4 in Supporting Information S1). Sediment samples were dried in an oven at 50°C and ground with a mortar and pestle to homogenize the material. Aliquots were made from each sediment sample for δ¹³C and δ¹⁵N bulk analysis and for δ³⁴S bulk analysis of select samples based on δ¹³C and δ¹⁵N results.

2.2 Stable Isotope Analyses

Isotope samples are run in labs at two different universities due to funding differences for the two research sites (MS and AL). The specific labs are indicated throughout the following section, and respective lab procedures, standards, and instrumental accuracy values are presented. All sediment stable isotope values are reported in standard delta notation (δ) as per mil (‰). Stable isotope values are reported based on international standards respective to each element, including Vienna Pee Dee Belemnite (VPDB) for δ¹³C values, atmospheric N₂ (air) for δ¹⁵N values, and Vienna Canyon Diablo Troilite for δ³⁴S values.

Samples from core 19OCS-SI39 were analyzed for carbon and nitrogen bulk stable isotope values (δ¹³C and δ¹⁵N) at the University of Louisiana at Lafayette Schubert Laboratory. Five samples from core 19OCS-SI39 were initially used to estimate carbon and nitrogen concentration ranges that were then used to determine target masses for all samples to be weighed for optimal detection on the instrument. Nitrogen data were collected on unacidified samples (after Brodie, Heaton, et al., 2011). The samples for organic carbon isotope analysis were weighed into silver capsules and acidified with 10% hydrochloric acid to remove carbonates, following the “capsule method” described in Brodie, Leng, et al. (2011). The acidified samples were dried overnight at 55°C and then wrapped tightly in tin capsules. Samples acidified for core 19OCS-SI39 did not produce gas bubbles and no carbonates were identified via microscopy prior to acidification, so the acid treatment is assumed to have negligible influence on the carbon isotope values. All samples were analyzed using a Thermo-Fisher Delta V Advantage Isotope Ratio Mass Spectrometer interfaced with a Flash EA 1112 Elemental Analyzer. Each of the carbon and nitrogen analyses has three in-lab reference materials measured together with the samples as calibration standards. Their isotope values range from −43.51‰ to −8.15‰ for carbon and −4.34‰ to 5.92‰ for nitrogen. A fourth in-lab reference material, JGLUC (glucose) for carbon (δ¹³C = −10.52‰) and JALA (alanine) for nitrogen (δ¹⁵N = −3.16‰), was also analyzed with the samples as unknown. Analytical precision based on the standard deviation (1σ) of repeated reference materials was ±0.32‰ for δ¹³C and ±0.24‰ for δ¹⁵N (Table S3 in Supporting Information S1).

Sediment samples from cores 15DF-1 and 15DF-3B were analyzed for carbon and nitrogen bulk stable isotope values (δ¹³C and δ¹⁵N) at the LSU Stable Isotope Ecology Laboratory. Seven samples from a range of percentages of organic carbon were first used to create a concentration calibration to help determine target weights for all samples. Following the development of the concentration calibration, all samples were weighed to their respective target masses into tin capsules, which were then wrapped tightly. The prepared samples were analyzed for stable isotope values of carbon and nitrogen as well as the total elemental abundance of carbon and nitrogen. Samples were flash combusted using a Costech ECS4010 Elemental Analyzer coupled to a Thermo-Fisher Delta Plus XP Continuous-Flow Stable Isotope Ratio Mass Spectrometer. Raw δ values were normalized using a two-point system of USGS-40 (δ¹³C = −26.40‰, δ¹⁵N = −4.50‰) and USGS-41 (δ¹³C = 37.60‰, δ¹⁵N = 47.60‰) glutamic acid standard and reference materials. A fish muscle (Red Drum; Sciaenops ocellatus) laboratory standard and glutamic acid standard reference materials (USGS-40 and USGS-41) were also used to assess instrument accuracy and precision. The analytical precision based on the standard deviation (1σ) of repeated reference materials was ±0.20‰ for δ¹³C and ±0.13‰ for δ¹⁵N (Table S3 in Supporting Information S1). Duplicate measurements for δ¹³C, δ¹⁵N, and C/N were conducted for a subset of samples (Table S4 in Supporting Information S1).

Samples from cores 19OCS-SI39, 15DF-1, and 15DF-3B were analyzed for bulk sulfur stable isotope values (δ³⁴S) at the LSU Stable Isotope Ecology Laboratory. Five samples from the top (0–20 cm) of each sampled core section were analyzed in 5-cm intervals, along with samples from the remainder of each core at 20-cm intervals. Samples were first used to establish a concentration correction and determine target masses. Samples for sulfur stable isotope analysis were weighed into tin capsules, and vanadium pentoxide (V₂O₅) was added as a catalyst for combustion. The mass of V₂O₅ added was roughly twice the mass of each sample to ensure full combustion. The prepared samples were flash combusted and run using a Thermo Scientific EA IsoLink Elemental Analyzer System coupled to a Thermo-Fisher Delta V Advantage Continuous-Flow Stable Isotope Ratio Mass Spectrometer. A certified reference material of known δ³⁴S value (Fish Gelatin; SKU: B2215) was measured after every seven samples and used to drift correct samples and reference materials using a linear model (Werner & Brand, 2001). Drift-corrected δ values were normalized using a two-point system of silver sulfide (IAEA-S2, δ³⁴S = 22.62‰ and IAEA-S3, δ³⁴S = −32.49‰) standard reference materials to assess accuracy and precision. A fish muscle (Red Drum; Sciaenops ocellatus) laboratory standard and silver sulfide standard reference materials were then used to assess analytical accuracy and precision. The analytical precision based on the standard deviation (1σ) of repeated analysis of reference materials was ±0.34‰ for δ³⁴S (Table S3 in Supporting Information S1).

2.3 Palynological Analyses: Pollen and Spores

Five sediment samples from core 19OCS-SI39 were sent to GeoLab Limited for palynological processing. Hydrochloric and hydrofluoric acid were used to remove carbonates and silicates, respectively, and one tablet of Lycopodium spores (batch #50220211, x = 18,407 spores) was added as an internal spike to calculate absolute concentrations of palynomorphs per gram of dried sediments (Stockmarr, 1971). The concentration is calculated using the equation determined by Benninghoff (1962):

$$$C=\frac{(\mathit{Pc}\times \mathit{Lt}\times T)}{(\mathit{Lc}\times W)}$$$ (1)

where C is the concentration (number of palynomorphs per gram of dried sediment), Pc is the number of each palynomorph counted, Lt is the number of Lycopodium spores per tablet, T is the total number of Lycopodium tablets added per sample, Lc is the number of Lycopodium spores counted, and W is the weight of dried sediment processed in grams.

All pollen and spores were counted and imaged at the LSU Center for Excellence in Palynology using an Olympus microscope BX43 under oil immersion objectives (60× and 100×) until 300 palynomorphs were observed. We referenced Crouch (2010) to assist with the identification of pollen specimens. Pollen concentrations per gram of dried sediment and percent relative abundances were calculated, and data were imported into TiliaIT software (Grimm, 1991) to analyze and visualize the results. Pollen results for cores 15DF-1 and 15DF-3B are reported in previous studies (Garretson, 2022, 2024; Reese et al., 2018a, 2018b) using similar methods.

2.4 Linear Discriminant Analysis

This study applies an additional analysis to the geochemical data via a 3-factor linear discriminant analysis (LDA) for environmental classification using the Modern Applied Statistics with S package in R software and subsequent data treatment methods (Venables & Ripley, 2002). The LDA considers relationships among δ¹³C, δ¹⁵N, and δ³⁴S values to define the linear boundaries among data from five environment types: terrestrial C₃ (TC3), terrestrial C4 (TC4), freshwater marsh (FM), saltwater marsh (SM), and open marine (OM), which were chosen because they are coastal ecosystems commonly found in the southeastern United States along the northeastern Gulf of Mexico. Isotope data (δ¹³C, δ¹⁵N, and δ³⁴S) for these environment types were compiled from 23 published studies to train the LDA model and identify isotopic signatures of the considered coastal environments (Table 1, Figure S7 and Table S6 in Supporting Information S1). Values in the possible range for δ¹³C and δ¹⁵N (50 in total), and δ³⁴S (25 in total) were identified, respective to each environment type. Fewer sulfur isotope values were identified for this analysis because δ³⁴S environmental studies are less common than those pertaining to δ¹³C and δ¹⁵N compositions. To reconcile this, the δ³⁴S values were averaged and extrapolated to establish a possible range of values for δ³⁴S (50 in total to match δ¹³C and δ¹⁵N). Isotope data generated as part of this study were then analyzed within the LDA model to produce estimations of the most probable environmental setting for the analyzed sediment intervals. Isotope results were only analyzed using LDA when a given sample had δ¹³C, δ¹⁵N, and δ³⁴S data; no interpolation or averaging was part of LDA model analysis and interpretation. These mathematical predictions are compared with the environment types as determined by the multiproxy environmental interpretations, which consider other variables in addition to stable isotope values (i.e., total elemental abundance, pollen, and spores). The summary values presented in Table 1 are not intended to be referenced as a whole for the paleoenvironmental interpretation of stable isotope data for individual isotope profiles (as each study referenced has respective limitations) but rather act as a simple representation of the data mined for the purpose of training the LDA (see Table S6 in Supporting Information S1 for the entire data set).

Table 1. Summary of Legacy Data Used to Train the Linear Discriminant Analysis (LDA) Model for Environmental Classification

Environment	δ^13C (‰ VPDB)		δ^15N (‰ air)		δ^34S (‰ VCDT)		References
	Mean	Range	Mean	Range	Mean	Range
Terrestrial C3 dominant (TC3)	−27.1	−32.4 to −22.6	3.55	−0.6 to 8.5	1.6	−1.3 to 4.5	1, 2, 6, 9, 17, 19
Terrestrial C4 dominant (TC4)	−13.6	−19.9 to −10.4	3.62	−3.21 to 12.4	4.3	−9.6 to 12.9	1, 2, 6, 8, 9, 10, 11, 20, 21
Freshwater marsh (FM)	−26.6	−32.2 to −16.3	7.7	−1.7 to 14	4.4	−3.4 to 18.3	1, 2, 3, 4, 5, 7, 12, 17
Saltwater marsh (SM)	−27.4	−29.4 to −23.6	10.6	2.9 to 16.5	12.8	4.6 to 20.0	3, 17, 22, 23
Open marine (OM)	−21.5	−26.0 to −9.0	5.5	1.7 to 9.9	20.8	18.8 to 22.1	2, 5, 7, 13, 14, 15, 16, 18

• Note. References for legacy data used in LDA include: ¹Wada et al. (1975), ²Meyers (1994), ³Chmura et al. (1987), ⁴Neill and Cornwell (1992), ⁵Khan et al. (2019), ⁶X. Wang et al. (2013), ⁷Gordon and Goñi (2003), ⁸Swap et al. (2004), ⁹Luo et al. (2018), ¹⁰Ma et al. (2012), ¹¹Szpak et al. (2013), ¹²Bergamino et al. (2014), ¹³White et al. (2007), ¹⁴Ruttenberg and Goni (1997), ¹⁵Ogrinc et al. (2005), ¹⁶Peters et al. (1978), ¹⁷Peterson and Howarth (1987), ¹⁸Gomes and Johnston (2017), ¹⁹Chukhrov et al. (1980), ²⁰Zhou et al. (2021), ²¹Currin et al. (1995), ²²Cloern et al. (2002), and ²³Kwak and Zedler (1997).
• Abbreviations: VCDT, Vienna Canyon Diablo Troilite; VPDB, Vienna Pee Dee Belemnite.

3 Results

Three sediment cores (19OCS-SI39, 15DF-1, and 15DF-3B) were analyzed for δ¹³C, δ¹⁵N, carbon to nitrogen ratio (C/N), δ³⁴S, and total sulfur (TS) compositions (Figure 2; Table S5 in Supporting Information S1). The bottom-most samples in core 19OCS-SI39 (445 and 450 cm) have δ¹³C values of −24.6‰ and −26.9‰, respectively, which are relatively ¹³C-rich compared with 11 out of the 16 samples analyzed for δ¹³C. Removing these samples from the total δ¹³C value average shifts the average by 0.5‰ to −28.3‰ ±0.97‰ (1σ). Core 19OCS-SI39 has a change in the sediment facies at 442 cm and therefore samples at 445 and 450 cm core depth represent a mixed erosional layer that is not elaborated on further in this study. The C/N values range from 16.4 to 23.3 except for two higher values at 395 cm (C/N = 29.4) and 415 cm (C/N = 39.0). In core 19OCS-SI39 from 450 to 395 cm, δ³⁴S values are elevated, with an average of 7.9‰ ± 1.3‰, 1σ (n = 4), and above 395 cm core depth, there was a ∼6.0‰ decline in δ³⁴S value to 0.15 ± 0.4‰, 1σ (n = 4).

Figure 2

Geochemical results for cores (a) 19OCS-SI39, (b) 15DF-1, and (c) 15DF-3B. Color images of the core sections are shown on the left with facies (brown text) and pollen assemblages (green text) noted (see Figures S5 and S6 in Supporting Information S1) (Garretson, 2022; Reese et al., 2018a, 2018b). Organic carbon (%), δ¹³C (‰ VPDB), δ¹⁵N (‰ Air), C/N, δ³⁴S (‰ Vienna Canyon Diablo Troilite), and total sulfur (%) are plotted for each core. Percent organic carbon was determined as part of the carbon and nitrogen analyses for core 19OCS-S139 and for cores 15DF-1 and 15DF-3B%, organic carbon and sediment descriptions are from DeLong et al. (2021) and Gonzalez et al. (2017). Error bars are analytical precision (1σ) (Tables S3 and S4 in Supporting Information S1) and if not shown are smaller than the symbol or not determined (for % organic carbon a conservative uncertainty for LOI measurements is 2% (Heiri et al., 2001)). Gray circles indicate the location of ¹⁴C dates for each core (Table S2 in Supporting Information S1 for core 19OCS-S139) and see Table 2 in DeLong et al. (2021) for cores 15DF-1 and 15DF-3B. The gray circles outlined in black are radiocarbon dead (i.e., >43.5 ka BP). Gray bars indicate the erosional layer in cores.

Core 15DF-1 has δ¹³C values that average −28.5‰ ± 0.8‰ (1σ) and are generally stable throughout the core until the uppermost samples (400 and 395 cm) increase to −25.8‰ and −27.4‰, respectively (Figure 2b). The C/N values range from 20.2 to 34.2 below 425-cm depth, then peak at 420-cm depth (C/N = 47.2) and then decline steadily to below mode values at the top of the core (C/N = 12.2 at 395 cm). In core 15DF-1, the δ³⁴S values were higher downcore from 475 to 455 cm, where the average δ³⁴S value is 18.3‰ ± 1.1‰, 1σ (n = 2). In the middle (435–415 cm) of the sampled section, the δ³⁴S values average 11.1‰ ± 0.6‰, 1σ (n = 2). The δ³⁴S values increase upcore (410–400 cm) and at 395 cm, the δ³⁴S value is 19.9‰. Core 15DF-1 has the highest TS values compared with cores 19OCS-SI39 and 15DF-3B, with TS values that range from 0.6% to 9.6%. The highest TS values occur from 415 to 405 cm with samples from this interval averaging 8.4%. All other samples from core 15DF-1 average 1.6% for TS.

Similar to core 15DF-1, core 15DF-3B has δ¹³C values that average −28.6‰ ± 0.5‰ (1σ) and are generally stable throughout the core (Figure 2c). Mode values in core 15DF-3B are not significantly different from core 15DF-1 and show a similar peak in C/N values in the upper half of the core (C/N = 40.0 at 315 cm) before declining at the core top (C/N = 16.2 at 300 cm). In core 15DF-3B, the δ³⁴S values increase moving upcore (440–305 cm). The greatest amounts of sulfur in this core occur at 360 cm (6.6%) and 315 cm (6.3%), the two of which average 6.4%. All other TS values from core 15DF-3B average 2.7%. The average TS for cores 15DF-1 and 15DF-3B combined is 3.7%, whereas the average TS in core 19OCS-SI39 is 1.8%.

Five sediment samples from core 19OCS-SI39 were analyzed for pollen and spores (Figure 3). Pollen assemblages in samples 3, 4, and 5 are arboreal-dominant with the highest pollen abundances including T. distichum (baldcypress), Pinus spp. (pine), and Quercus spp. (oak); samples 3, 4, and 5 each have <1% of aquatic vegetation. In sample 2, Typha spp. is found abundantly in the sample. Typha vegetation typically includes freshwater dominant grasses (i.e., Typha latifolia, also known as cattail). Pollen assemblages in sample 1 reveal less Typha spp. relative to sample 2; however, herbaceous plants such as Amaranthaceae, Ambrosia, and Asteraceae become more established upcore in samples 1 and 2.

Figure 3

Relative abundance (%) of palynomorphs and concentration per gram of dried sediment relative to core depth for core 19OCS-SI39 (Fontenot et al., 2024b). A total of five processed and analyzed samples are reported. Vegetation classified as arboreal (brown), aquatic-arboreal (purple), non-arboreal (green), and aquatic non-arboreal (blue) is indicated at the bottom of the figure.

4 Discussion

4.1 Mississippi Site Environmental Interpretation

The combined findings from geochemical and palynological analyses for core 19OCS-SI39 reveal that during the early Holocene, the MS site was well inland (Figure 1) as no marine or brackish component was observed with either analytical technique. The combined assessment of total elemental abundance and stable isotope values of sediments from core 19OCS-SI39 indicates a swamp ecosystem that shifts to an FM (Figure 2). The geochemical composition of core 19OCS-SI39 provides insight into the ecological and climatic conditions of the Gulf Coast during the early Holocene (∼10–11 ka). The δ¹³C values throughout core 19OCS-SI39 average −27.8‰ and are therefore characteristic of C₃-dominant vegetation, which typically ranges from −27‰ to −25‰ (Degens, 1969; O'Leary, 1988). Similarly, the δ¹⁵N values of samples from the MS site indicate the presence of terrestrial soils, which commonly range from 2‰ to 5‰ (Broadbent et al., 1980). The δ³⁴S values in the 450 to 395 cm interval are consistent with environments that favor terrestrial and aquatic plants, typically between −3‰ and 7‰ (Krouse & Grinenko, 1991). Both the C/N and TS values consistently remain stable, suggesting non-marine conditions during the deposition of these sediments (Gordon & Goñi, 2003).

Terrestrial plants sequester carbon primarily from atmospheric carbon dioxide (CO₂). Variation in atmospheric δ¹³C conditions across different temporal intervals influences the overall δ¹³C composition in plants. For example, industrialization has caused an increase in the amount of atmospheric CO₂, and a subsequent lowering of δ¹³C values in the atmosphere (δ¹³C_atm) as described by the “carbon-13 Suess effect” (C. D. Keeling, 1979). During the late Pleistocene and early Holocene intervals pertinent to this study, δ¹³C_atm is estimated to be around −6.5‰ (Bauska et al., 2016; Eggleston et al., 2016) and CO₂ ranges from 180 to 270 ppm, whereas modern (21st century) atmosphere has δ¹³C_atm < −8.1‰ and CO₂ > 370 ppm (R. F. Keeling et al., 2017). Overall changes in δ¹³C_atm and CO₂ across different temporal intervals must therefore be recognized when assessing δ¹³C compositions as a tracer of past vegetative conditions (Schubert & Jahren, 2012). Thus, average δ¹³C values determined in this study for the late Pleistocene and early Holocene (i.e., −28‰ to −30‰) are equivalent to modern-day values less than −32‰ (Schubert & Jahren, 2015). Such low values in a modern context suggest, downcore, that the MS deposits resemble a closed canopy environment (Kohn, 2010), such as a swamp, where ¹³C depleted CO₂ is recycled within the system.

The isotopic profiles of the δ¹³C, δ¹⁵N, and δ³⁴S values at the MS site exhibit a shift starting at ∼405 cm moving upcore, suggesting an onset of ecosystem change at the time of deposition and possibly a climate shift. At this interval, ¹³C and ¹⁵N are enriched and ³⁴S is depleted, thus revealing a change in the ecosystem, increased nutrient supply, and a transition from a more reducing to a more oxidizing environment (Farquhar et al., 1989; Rosenbauer et al., 2009; Thode, 1991). Collectively, the observed trends are interpreted as freshwater inundation of the forest ecosystem and a shift from a closed canopy swamp to a more open marsh environment, thus further reflecting the ephemeral and inland nature of this past ecosystem (Figure 1). Stability in the C/N values for this interval further confirms that the observed environmental shift is not due to additional marine influence, which would manifest as a steady decrease in the C/N ratio (Gordon & Goñi, 2003). The source of freshwater may be the result of increased runoff via precipitation or changes in fluvial input. Given the transition from layered peat deposits to fine-grained humic mud at this interval, the facies change (Miller et al., 2021) (Figure S1 in Supporting Information S1) and subsequent shifts in the geochemical data are potentially representative of enhanced freshwater deposition.

The palynological analyses of samples from core 19OCS-SI39 (Figure 3) agree with the environmental interpretations as determined by the geochemical analyses (Figure 2). Pollen and spore assemblages downcore in samples 3, 4, and 5, spanning from 418 to 447 cm, reveal arboreal-dominant vegetation in the region with the most abundant taxa being T.distichum (baldcypress), Pinus spp. (pine), and Quercus spp. (oak), all of which are common types of trees found regionally in swamp environments on the northern Gulf Coast (Battaglia et al., 2012). Upcore, the palynomorph assemblage in sample 2 reveals an abundance of Typha pollen grains. This genus includes FM plants such as Typha latifolia (cattail). Additional herbaceous plants are established in samples 1 and 2, which have greater abundances of Ambrosia (ragweed), Amaranthaceae, and Poaceae (grasses) compared with samples analyzed downcore. This shift in the palynomorph data could be the result of ecological succession following environmental change at the site or a shift upgradient that is observed at the study site due to the eolian and hydrologic transport of palynomorphs. However, the shift observed in the in situ geochemical data suggests the former; thus, an FM is likely established. The interpreted freshwater inundation of the swamp around 405 cm from geochemical markers is consistent with the vegetative compositions shifting to include more grasses and is especially reflected by the increase in Typha observed in sample 2 (404–407 cm).

In summary, the results of these analyses support the presence of inland freshwater ecosystems at the MS site in the northeastern Gulf Coast region during the early Holocene when the sea level was still rising, despite an overall climate state that was cooler compared with modern conditions (Hansen et al., 2013). The enhanced freshwater mud deposition at the top of the sampled core section would have acted as a mechanism of rapid burial, thus physically preserving the peat deposits downcore and promoting the establishment of marsh conditions.

4.2 Alabama Site Environmental Interpretation

The combined geochemical and palynological analyses of sediments from cores 15DF-1 and 15DF-3B (Figure 2) reveal a terrestrial swamp well inland (Figure 1b) that transitions to a brackish marsh transitional environment during the late Pleistocene that later becomes fully marine in the Holocene in the interbedded sand and mud facies and Holocene sand (Figures S2–S4 in Supporting Information S1). The δ¹³C values of cores 15DF-1 and 15DF-3B indicate C₃-dominant vegetation at the time of deposition for the terrestrial sediments (Degens, 1969; O'Leary, 1988). The δ¹⁵N values in both cores are similarly representative of terrestrial plants and soils (Broadbent et al., 1980; Fry, 1991; Pate et al., 1994). This interpretation is further supported by stable C/N values from 475 to 425 cm in core 15DF-1, which implies non-marine conditions (Gordon & Goñi, 2003). Upcore in core 15DF-1, a transitional phase occurs in multiple geochemical variables, suggesting the onset of ecosystem change. The δ¹³C and δ¹⁵N values increase and the C/N steadily decreases. The intensified range of fluctuations of δ¹⁵N values in core 15DF-3B moving upcore signifies nutrient flux in the environment. This is further evidenced by the general increase in δ³⁴S and TS in core 15DF-3B as well as the decrease in C/N values in the upper portion of core 15DF-3B that was examined for this study. The C/N values remain stable until the uppermost core section of 15DF-3B, after which a steady decrease in C/N is observed from 330 to 315 cm, signifying the onset of marine influence and associated increases in salinity (Gordon & Goñi, 2003). Collectively, these observations reveal a preserved progression from a terrestrial wetland forest to a brackish or marine environment in cores 15DF-1 and 15DF-3B.

These interpretations are supported by previously reported palynological assemblage analyses of cores 15DF-1 (Reese et al., 2018a) and 15DF-3B (Garretson, 2022) (Figures S5 and S6 in Supporting Information S1). The study of Reese et al. (2018a) identifies five major environmental zones in core 15DF-1 from the bottom of the core to the top: Cypress/Tupelo forest community (455–480 cm), open grass-dominated marsh (435–450 cm), Cypress/Alnus forest community (420–430 cm), open grass-dominated marsh (405–415 cm), and marine (0–400 cm). The current study reports geochemical values from the same core interval as the study of Reese et al. (2018a) (Figure 2b, Figure S2 in Supporting Information S1); however, our geochemical results indicate the presence of three major environmental zones since geochemistry cannot distinguish differences in forest tree species (i.e., Cypress/Tupelo and Cypress/Alnus). The different analytical perspectives agree for interpretations regarding the two most recent depositional environments (open marsh and marine). The analyses diverge in interpretation for the lower portion of core 15DF-1 (420–475 cm). At this interval, Reese et al. (2018a) identified three distinct environmental zones, whereas the geochemical compositions presented in this study remained consistent and failed to capture the finer variability from the palynomorph assemblages. For this reason, it is suggested that the transitions from closed forest to open grass marsh are geochemically more distinct than the overall terrestrial forested swamp environments observed downcore. Furthermore, the geochemical results suggest that the open marsh community (4.35–4.50 m; Figure 2b, Figure S5 in Supporting Information S1) between the wetland forest communities is freshwater and not saltwater.

The study of Garretson (2022) conducted a palynological assessment on samples from core 15DF-3B in the same section analyzed in this study and identified a transitional boundary from baldcypress swamp to open marsh environments (Figure 2c, Figure S6 in Supporting Information S1). These findings agree with the geochemical assessment of paleoenvironmental conditions at this site, further supporting that the sediments from core 15DF-3B were initially deposited in a swamp environment and later transitioned to marsh conditions. Geochemical analyses from this study further refine the interpretation based on palynomorph assemblages by revealing that the aforementioned marsh environment upcore was likely a brackish or SM given the elevated δ³⁴S values. At 300 cm, the δ³⁴S value is 1.6‰, a ∼16‰ decrease from the previously sampled core section at 305 cm. This shift indicates terrestrial freshwater conditions that do not agree with interpretations based on δ¹³C, δ¹⁵N, C/N, and palynomorph interpretations by Garretson (2022). Possible sources of this discrepancy include a problem in the sample preparation that caused the depletion of ³⁴S, and full combustion of the sample not occurring in the isotope ratio mass spectrometer among others.

The geochemical variability of sulfur in core 15DF-1 is surprising in the context of similar analyses from other studies and is partially inconsistent with the other geochemical results in this study. Stable isotope values of sulfur from 455 to 475 cm average 18.3‰, indicating near-marine conditions (Figure 2b) (Aharon & Fu, 2003). The presence of marine conditions at this interval contradicts the other geochemical interpretations and palynological assemblages from Reese et al. (2018a) as well as sedimentological results and the presence of tree stumps in this facies (DeLong et al., 2021; Gonzalez et al., 2017). Based on the ¹⁴C dates and OSL with their large uncertainties (DeLong et al., 2021), the AL site when the trees were alive could have been between 0 and 80 m above sea level and at most 60–100 m from the coastline (Figure 1b) assuming modern bathymetry is a good proxy for past coastline and global sea level is representative for our site. As a result, these values likely do not represent true marine conditions but rather are chemically linked to a different aspect of general sulfuric cycling or possibly diagenesis. Other possible sources of sulfur fractionation leading to changes in ³⁴S enrichment include the presence of sulfate-reducing bacteria, volcanic activity, and chemosynthetic oxidation, among others (Shen et al., 2022; Thode, 1991). Aside from this anomalous interval, the sulfur isotope values thereafter begin to agree with interpretations based on carbon, nitrogen, and palynological analyses. Starting at 435 cm and continuing upcore, the δ³⁴S values transition sequentially from ∼10‰ to ∼15‰ in alignment with the environmental shifts from a forested swamp to an open marsh, which is likely a brackish or SM given the enriched ³⁴S. The sample at 395 cm has a δ³⁴S value of 19.9‰, representing fully marine conditions (Aharon & Fu, 2003). The sample at 395 cm is from the interbedded sand and mud facies of core 15DF-1 (Figure 2b; Figure S2 in Supporting Information S1), which is a mixture of Pleistocene and Holocene-aged material (DeLong et al., 2021; Gonzalez et al., 2017). Sediment samples from 325 to 400 cm were examined for pollen by Reese et al. (2018a); none were found suggesting a marine environment. Furthermore, foraminifera were found in this interval, which have a radiocarbon age of ∼3.9 cal ka BP at 330 cm down core (DeLong et al., 2021). In short, this upper interval reflects trends in δ³⁴S values due to an increasing salinity gradient (Fry & Chumchal, 2011).

Cores 15DF-1 and 15DF-3B were both recovered from the Alabama Underwater Forest. Though the recovered material in each sediment core cannot be confirmed as chrono-stratigraphically connected due to large dating uncertainties associated with OSL methods, DeLong et al. (2021) infer stratigraphic connections among the AL cores based on lithology and a common unconformity boundary (Figure S4 in Supporting Information S1). Both of the AL core profiles reveal that the terrestrial deposits are an inland swamp that transitions to an SM environment (Figures 2b and 2c). These shifts seen in geochemistry and in polymorphs could be associated with the rise and fall of sea level during MIS 3–5a (e.g., the rise from 65 to 60 ka or the fall from 75 to 70 ka; Figure 1), which are poorly understood for the Gulf of Mexico in this time interval. For example, (referring to Figure 1b), at 80 ka, the site may have looked like −20 m, at 65 ka, like −80 m, and at 60 ka like −40 m, thus experiencing a ∼40 m rise in sea level in ∼5,000 years. In contrast, the MS site is reflective of a swamp to FM environment (Figure 2). Despite these differences in depositional setting for the two sites, all three sediment cores have some level of similarity among geochemical compositions.

4.3 Linear Discriminant Analysis for Environmental Classification

Multivariate mathematical analyses can provide helpful insight into paleoenvironmental research questions because they are able to elucidate underlying trends and patterns in any given data set. Furthermore, LDA allows for determining the probability of each classification in order to assess the strength of the multi-proxy joint interpretation. For this reason, numerous multivariate methods are common in paleoenvironmental studies, including principal component (e.g., McCloskey et al., 2018), correspondence (e.g., Toledo et al., 2009), and linear or quadratic discriminant analyses (e.g., Alexandrino et al., 2017; Khan et al., 2019). Discriminant analysis, a form of supervised machine learning, is of particular interest because it allows for the classification of unknown elements in a data set into specified groups, such as environment type (Park, 1974). This is done mathematically by maximally separating data into groups based on underlying relationships among the variables considered. Despite this potential, the application of discriminant analysis methods considering stable isotopes for paleoenvironmental classification is lacking in the literature. For this reason, we establish a three-factor LDA model (variables considered are δ¹³C, δ¹⁵N, and δ³⁴S) that was trained with a composite data set of previously reported isotope values of known environment types from 23 papers (Table 1 and Table S6 in Supporting Information S1). The environment types considered by the LDA model are TC3, TC4, FM, SM, and OM. We then analyze the δ¹³C, δ¹⁵N, and δ³⁴S data for cores 19OCS-SI39, 15DF-1, and 15DF-3B within the trained LDA model to determine probable environment types for each sampled interval (Table 2). Since environmental groupings based on the model only consider the simultaneous relationship among δ¹³C, δ¹⁵N, and δ³⁴S variables, the LDA results are assessed against paleoenvironmental interpretations made based on individually resolved δ¹³C, δ¹⁵N, and δ³⁴S results for each sediment core alongside C, N, and S elemental values and palynomorph results as described in Sections 4.1 and 4.2.

Table 2. Environmental Classifications by Linear Discriminant Analysis of Stable Isotope Data

Core ID	Depth from core top (cm)	1st predicted environment	Probability (%)	2nd predicted environment	Probability (%)	Residual influence (%)
19OCS-SI39	375	TC3	88.4	FM	11.6	0
	380	TC3	88.3	FM	11.7	0
	385	TC3	86.7	FM	13.2	<1
	390	TC3	92.6	FM	7.4	0
	395	TC3	82.9	FM	16.7	<1
	415	TC3	83.0	FM	16.4	<1
	435	TC3	77.1	FM	21.7	1.2
	450	TC3	57.0	FM	37	6
15DF-1	395	OM	90.5	SM	9.1	<1
	400	OM	47.2	SM	39.1	13.7
	405	SM	30.7	TC3	28.8	40.5
	410	TC3	47.3	SM	21.6	31.1
	415	TC3	81.1	FM	16.7	2.2
	435	TC3	72.9	FM	21.6	5.5
	455	OM	66.9	SM	30.3	2.8
	475	OM	70.1	SM	22.7	7.2
15DF-3B	300	TC3	87.3	FM	12.7	0
	305	SM	45.3	OM	31.1	23.6
	310	TC3	33.9	SM	26.3	39.8
	315	TC3	75.8	FM	19.5	4.7
	320	TC3	76.1	FM	9.7	14.2
	340	SM	42.0	OM	40.9	17.1
	360	TC3	64.4	FM	25.8	9.8
	380	TC3	58.2	FM	30.3	11.5
	400	TC3	91.5	FM	8.4	<1
	420	TC3	89.6	FM	10.3	<1
	440	TC3	72.0	FM	25.4	2.6

• Note. Environment types are indicated by color: open marine (OM; blue), saltwater marsh (SM; tan), terrestrial C₃ (TC3; dark green), and freshwater marsh (FM; light green). No occurrences of terrestrial C₄ (TC4) environments were determined by the model for samples from the MS and AL deposits.

The LDA results, via consideration of only δ¹³C, δ¹⁵N, and δ³⁴S data, classify the MS sediments as most likely TC3 with an average predictive confidence of 82% (n = 8), followed by FM (Table 2). These classifications agree with the environmental interpretations of the geochemical and palynological analyses of samples from core 19OCS-SI39, which we assessed are representative of a swamp and FM (Figures 2a and 3). The LDA results further support our interpretation (see Section 4.1) that during the early Holocene, the MS site was an inland C₃-dominant freshwater ecosystem.

Similarly, the LDA results for core 15DF-3B classify mainly TC3 and SM (67% predictive confidence; n = 11), followed by primarily FM (Table 2). The LDA captures an environmental signal that agrees with interpretations from geochemical and palynological analyses for core 15DF-3B (see Section 4.2). However, the LDA fails to fully resolve the distinct transition toward an SM environment based on carbon, nitrogen, and sulfur isotopes alone as evidenced by combined results of geochemical, sedimentological, and palynological analyses. Furthermore, the TC3 geochemical signal of swamp and marsh environments in core 15DF-3B presides over the SM signal. In contrast, the highest LDA classification probabilities for core 15DF-1 from the AL site are identified as a combination of TC3, SM, and OM (63% predictive confidence; n = 8), followed secondarily by SM, FM, and TC3.

Samples downcore in core 15DF-1 were predicted by the LDA as mainly OM, which is not supported by other environmental interpretations using geochemical, sedimentological, and palynological analyses (see Section 4.2). Palynological interpretations from Reese et al. (2018a, 2018b) identify this depositional layer as a Cypress/Tupelo Community, which are freshwater swamps not typically located on the coast but inland because baldcypress is not salt tolerant (Middleton, 2016) (Figure 2b). Additionally, this portion of the core is a peat deposit with high percentages of organic carbon and woody debris throughout (Figures S2 and S3 in Supporting Information S1), which is not representative of marine environments from a sedimentological perspective, not to mention the large tree stumps found in this facies (DeLong et al., 2021). This site was located inland when the sediments were deposited between ~41 and 80 ka when the sea level was lower by 20–80 m from today (Figure 1). In this case, the LDA is likely considering the sulfur stable isotope values in this portion of core 15DF-1 as solely representative of saline conditions; therefore, terrestrial versus marine conditions in general. This discrepancy is not surprising as similar limitations with some of the sulfur isotopes in core 15DF-1 were previously elaborated on (see Section 4.2) as being potentially chemically linked to an aspect of sulfuric cycling other than the marine environment.

Overall, the confidence of the predictions for core 19OCS-SI39 is higher than the predictive confidence for cores 15DF-1 and 15DF-3B, possibly due to the differences in environment type for the MS and AL sites. Swamp ecosystems are often highly anoxic, even more so than marshes. These anoxic environments can produce authigenic material rich in iron sulfide via sulfate-reducing bacteria when sulfur-rich waters are present (McCloskey et al., 2018). This process is skewing the environmental classification by LDA at the AL site because the LDA, based on the training data set used, is attributing sulfuric values solely to salinity and not considering other factors such as biogeochemical reactions like sulfate reduction that are typical in anoxic, and especially euxinic, environments, nor the palynology and sedimentology results. Other studies have found similar phenomena when using geochemical analyses of trace element concentrations to classify freshwater swamp environments (e.g., McCloskey et al., 2018). We suggest this process is also reflected isotopically, though more studies are needed to fully understand how isotope ratios are represented in anoxic and euxinic environments, such as freshwater swamp environments, given that isotope data (particularly δ³⁴S) are lacking for these types of ecosystems. A post-depositional alteration may alter the δ³⁴S signal if saltwater intrusion occurs (McCloskey et al., 2018). In core 15DF-1 for the interval studied (390–480 cm) dated to 42–72 ka, saltwater intrusion did not occur for at least 30 ka after deposition (more likely 60 ka). Thus sulfur-driven diagenesis likely did not occur because the core location was inland and above sea level at the time of deposition until marine inundation ∼10 ka (Figure 1). Furthermore, the higher δ³⁴S values are below the lower δ³⁴S in the core (Figure 2b); therefore, diagenesis from saltwater intrusion would have impacted the younger sediments first. Once the site goes into the marine environment ∼10 ka, post-depositional alteration of the δ³⁴S signal could happen, but this is unlikely since the sediments from organic material and wood had already been preserved for ∼40–60 ka and had 4 m of sand covering the terrestrial deposits. Thus, we interpret this section as anoxic with sulfate-reducing bacteria driving the δ³⁴S signal.

The multiproxy data sets (geochemical and palynological) presented in this study (Figures 2 and 3, Figures S5–S6 in Supporting Information S1) operate as a check against LDA model results. In some cases, there is an agreement between the physical evidence and the predicted environment based on LDA, and at other times there is disagreement. This may be due to a lack of defined linear trend separations among the five environment types within the three-factor model or under-constrained data mining during the compilation of the training data set so that more training data could improve the LDA model. The LDA model established for this study attributes ∼97% of environmental variability between environments to the carbon and sulfur isotopes, meaning that the nitrogen isotopes only account for ∼3% of linear groupings (δ¹⁵N has a standardized score of ∼3). This suggests that nitrogen potentially lacks utility in contributing majorly to these paleoenvironmental classifications using LDA based on the data values and resolution considered in this study.

4.4 Contextualizing Gulf of Mexico Paleoenvironmental Sites

The combined geochemical, palynological, and multivariate analyses of bulk sediment samples from two northeastern Gulf of Mexico sites indicate swamp and marsh conditions during the late Pleistocene and early Holocene. Ecosystems such as these are anoxic and/or euxinic; thus, the associated biogeochemical properties promote the sustained preservation of organic material (Canfield, 1989). We suggest that this is an additional reason for the preservation of the terrestrial deposits, especially given that the sites are from two distinct temporal intervals and the lithostratigraphic units are time-independent of one another. Samples from cores 19OCS-SI39, 15DF-1, and 15DF-3B all contain wood, pollen, and peat with no evidence of alteration or previous shelf exposure, confirming the hypothesis that innate biogeochemical processes of swamp and marsh environments aided in the preservation of these paleo-terrestrial deposits.

Though the environmental interpretations are different by site—the MS site is interpreted as a swamp transitioning to FM environment, whereas the AL site is interpreted as a swamp transitioning to SM environment—all are representative of anoxic (possibly euxinic) environment types commonly found in the Gulf Coastal Plain environment today (Ward & Tunnell, 2017). During the late Pleistocene and early Holocene intervals, temperatures were cooler, and sea level extended farther offshore compared with modern times (Figure 1; Hansen et al., 2013). Despite these differences in overall climatic conditions, we find that Gulf Coastal Plain environments during the late Pleistocene and early Holocene are not so different ecologically from those presently found in this same region, potentially due to glacial conditions being concentrated in the higher latitudes and polar regions. Foraminiferal temperature reconstructions from Gulf of Mexico marine sediment cores suggest that conditions in the Gulf of Mexico are inconsistent with Greenland atmospheric temperatures during late Pleistocene intervals of glacial advance (Hill et al., 2006) and instead likely follow climate change trends that are similar to what is observed in the southern hemisphere (Rohling et al., 2008). For this reason, it is difficult to fully understand changes in the ecology of Gulf Coastal Plain ecosystems because the climate in this region is distinct from the climate trends observed throughout much of the northern hemisphere during the same temporal interval.

The northern Gulf of Mexico is in the subtropical zone and is part of the Western Hemisphere Warm Pool (C. Wang & Enfield, 2001). Here, the Caribbean Current brings warm tropical waters northward that are circulated into the Gulf of Mexico via the Loop Current that eventually forms the Gulf Stream. This circulation system promotes a warm subtropical environment. During the cooler intervals of glacial advancement, this circulation of warm water potentially mitigated some of the more extreme impacts of northern hemispheric climate variability in this region, thus allowing swamp and marsh environments, as reflected by analyses in this study, to persist. This is not to say that climate conditions during glacial advances do not impact the Gulf of Mexico, as evidenced by abrupt events such as changes in sea level (Donoghue, 2011), the numerous meltwater pulse events that resulted from Laurentide Ice Sheet meltwater entering the Gulf (Ferguson et al., 2018; Kehew & Teller, 1994; Knox, 1996) and evidence of Heinrich events as found in pollen records from the Florida Peninsula (Grimm et al., 1993). Rather, the snapshot perspectives offered in this study imply that Gulf Coastal Plain ecologic compositions have broadly persisted during the late Pleistocene and early Holocene despite geologically short-term changes caused by these abrupt events.

Past environmental conditions are preserved in multiple ways, such as in the geochemical compositions of deposited material, evidence of micro- and macrofossils, and lithologic and mineralogic features. It is possible to observe ecologic changes from bulk stable isotope analyses (Byrne et al., 2001), but this only provides a limited perspective because a bulk sampling scheme is poorly resolved. To account for this weakness, the continual development of multiproxy records that incorporate geochemical, micropaleontological, lithological, and mineralogical proxies is necessary to investigate past environmental and ecological changes. Unfortunately, robust data are often limited, and a multiproxy perspective cannot always be attained due to the limited availability of sample material, diagenesis, or alteration within the archive, or reworking of material in the strata. For this reason, it is crucial to continue developing other tools that can provide additional insight into paleoenvironmental interpretations that do not require such an extensive multiproxy approach. For example, the use of geochemical methods that capture greater resolved variability than bulk sampling (i.e., compound or position-specific stable isotopic analyses (Brenna, 2001; Rossmann et al., 1991)) or the application of computational techniques (e.g., mixing models, machine learning, etc.). To continue specifically developing the analyses of these data, proxy analysis of cores elsewhere in the northern Gulf Coast is needed to improve the confidence of the existing interpretations. Additional analyses on past euxinic environments in freshwater and coastal settings are required to refine our understanding of how biogeochemical reactions innate to those ecosystems might influence geochemical data, especially from a stable isotopic perspective. In doing so, anoxic and/or euxinic environments such as swamps and marshes can be separated into distinct environmental categories, which help refine environmental classifications based on multiple stable isotope ratios. Though few studies to date have constructed a sufficiently robust analysis of swamp environments to be effective in the context of this study, the Gulf Coastal Plain and northern Gulf of Mexico have proven to be a candidate region to conduct additional investigations on similar past environments. The preservation of such terrestrial deposits in the marine environment can be attributed in part to the anoxic and/or euxinic nature of swamp and marsh ecosystems (Canfield, 1994; Ingall et al., 1993; Pratt, 1984). Seeking out new locations with similarly preserved deposits will aid in efforts to understand the responses of these environments to climatic change and how these environments are preserved through time.

5 Conclusions

This study reveals that late Pleistocene and early Holocene age coastal paleoenvironments located today south of Mississippi and Alabama are characterized as swamp and marsh environments. These sulfur-rich, anoxic environments promote the formation of authigenic material rich in iron sulfide, thus aiding in the preservation of organic compounds. In addition to the biogeochemical conditions within the deposited material, the mechanism of burial, which promoted the rapid accumulation of soft sediments throughout the northern Gulf of Mexico region, also aided in the remarkable preservation of these terrestrial deposits. Altogether, this study provides data toward understanding how coastal environments have evolved since the late Pleistocene and reveals the factors that aid in the preservation of such valuable natural archives. Marine sites with preserved terrestrial coastal sediments, like the ones in this study, are valuable analogs that advance our understanding of the resiliency of Gulf Coastal Plain wetlands to climate change. Future investigations should focus on environments that promote the preservation of past terrestrial deposits via the following factors: (a) are at topographically lower elevations relative to modern sea level, (b) have the accommodation space to allow for rapid burial, (c) are near sources with regular sedimentary deposition such as fluvial features, (d) are on passive tectonic margins, and (e) are anoxic or euxinic. Possible ideal candidate regions in addition to the northeastern Gulf of Mexico that meet the suggested criteria for preservation are the southern Bangladesh coastal area in the Bay of Bengal (Bomer et al., 2019; Wilson et al., 2017), the maritime continent of southeast Asia, and the peatland coastal region of northern Scotland and Doggerland in the North Sea (Charman et al., 2006).

Acknowledgments

Study collaboration and funding were provided by the US Department of the Interior, Bureau of Ocean Energy Management (BOEM), Environmental Studies Program, Washington, DC under Agreement Numbers M20AC10002 and M15AC00016. Additionally, core data acquired through BOEM Agreement Number M16AC00012 were used in this study. We would also like to thank the Wallace Research Foundation for providing initial funding. Additional research costs were funded in part by the LSU Department of Geography and Anthropology and the Association of the Marine Laboratories of the Caribbean. We thank the Field Support Group of the Coastal Studies Institute at LSU for their assistance in field operations. We thank Sam Bentley, Suyapa Gonzalez, and Jonathan Truong at LSU for their contributions to the Alabama Underwater Forest project and sediment core analyses. A special thank you to Tom Guilderson of the Center for Accelerator Mass Spectrometry, Lawrence Livermore National Laboratory for his help with radiocarbon dating analysis and interpretations. We are thankful to Yingfeng Xu of the University of Louisiana at Lafayette and to Angela Stahl and Vann Smith of LSU for assistance with analyses. We deeply thank Ben Raines for his local knowledge of the site and support of our investigation of the Underwater Forest. This report has been technically reviewed by BOEM, and it has been approved for publication. The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the opinions or policies of the Bureau of Ocean Energy Management. Mention of trade, firm, or product names does not constitute their endorsement or recommendation for use by the U.S. Government.

Open Research

Data Availability Statement

Core metadata and International Generic Sample Number (IGSN) are archived at System for Earth Sample Registration (www.geosamples.org; core 15DF1 https://doi.org/10.58052/IEADF151A, core 15DF3B https://doi.org/10.58052/IEADF153B, and core 19OCS-SI39 https://doi.org/10.58052/IE1930001). The geochemical, and palynomorph data that support the findings of this study are openly available at the National Centers for Environmental Information World Data Service for Paleoclimatology 325 Broadway, Boulder, Colorado; IGBP PAGES/World Data Center for Paleoclimatology Data Contribution Series #37260 (Fontenot et al., 2024a, https://www.ncei.noaa.gov/access/paleo-search/study/37260) and Neotoma Paleoecology Database https://data.neotomadb.org/58806 (Fontenot et al., 2024b). The LDA code in R and training data are provided in Supporting Information S1 and Supporting Data Sets S1–S3. The exact study site location coordinates are not publicly available due to privacy or ethical restrictions to protect the sites.