The Distribution Of Contemporary Saltmarsh Foraminifera In A Macrotidal Estuary: An Assessment Of Their Viability For Sea-Level Studies

¹School of Environmental Sciences, Department of Geography, University of Liverpool, Roxby Building, Liverpool, Merseyside L69 7ZT, UK

²School of Natural Sciences and Psychology, Department of Geography, Liverpool John Moores University, Byrom Street, Liverpool, Merseyside, L3 3AF, UK

³National Oceanography Centre, Joseph Proudman Building, 6 Brownlow Street, Liverpool, Merseyside L3 5DA, UK

*Corresponding Author:

Hayley Mills
British Oceanographic Data Centre
Joseph Proudman Building
6 Brownlow Street, Liverpool
L3 5DA, UK
Tel: +44-0-151 7954974
E-mail: haymil@bodc.ac.uk

Received date: June 24, 2013; Accepted date: July 24, 2013; Published date: July 26, 2013

Citation: Mills H, Kirby J, Holgate S, Plater A (2013) The Distribution of Contemporary Saltmarsh Foraminifera in a Macrotidal Estuary: an Assessment of Their Viability for Sea-Level Studies. J Ecosys Ecograph 3:131. doi:10.4172/2157-7625.1000131

Copyright: © 2013 Mills H, et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and and source are credited.

Visit for more related articles at Journal of Ecosystem & Ecography

Abstract

An analysis of dead benthic foraminiferal assemblages and associated environmental variables is presented to establish the controls on species distribution and zonation on two macrotidal saltmarshes in the Mersey estuary with the aim of evaluating the use of foraminifera in reconstructing past sea levels. The combined results from five contemporary transects from two sites illustrate that where there is a sufficient elevational range, foraminifera distribution and zonation shows a good relationship with altitude, with a high-to-middle marsh zone characterised by Haplophragmoides wilberti , Jadammina macrescens, and Miliammina fusca, and a low marsh zone composed of similar agglutinated species with calcareous species including Brizalina spp., Elphidium spp., and Haynesina spp. Cluster analysis and partial Redundancy Analysis (pRDA) reveals that the elevational control decreases with respect to other environmental variable inter-correlations when the sampling elevation range is reduced. It is proposed that this is a key issue for macrotidal environments when the contemporary sampling range represents a small proportion of the spring tidal range (i.e. <10%). Limiting the contemporary dataset to agglutinated species only, a tidal level transfer function can be obtained which offers acceptable predictability and precision (r2jack = 0.79, RMSEPjack = 0.13 m) for the reconstruction of past sea level.

Keywords

Foraminifera; Elevation transfer function; Mersey estuary; Sea level; Saltmarsh; Macrotidal estuary

Introduction

Sea-level trends from instrumental data have the potential to be extended back in time using geological and micropalaeontological evidence (e.g. foraminifera, diatoms and testate amoebae) obtained from saltmarsh sediments from relatively pristine to highly industrialised locations [1-3]. This enables our present understanding of sealevel changes to be put into a longer- term context [4]. In particular, foraminifera preserved within saltmarsh sediments have been utilised in many sea-level studies due to their strong and quantifiable relationship with elevation within the tidal frame [5-9].

Foraminifera-based sea-level reconstructions are founded on the present-day relationship between their modern distribution and the tidal frame. Saltmarsh foraminiferal assemblages may be controlled by a number of variables (e.g. salinity, temperature, grain size, etc.) that may have no direct relationship to elevation in the tidal frame [10-13]. Therefore the significance of elevation as a primary control on species distribution must be tested to determine whether the modern dataset is appropriate to use for sea-level reconstruction [5,14-17]. In addition, several studies find that sea-level reconstructions based upon local calibration of the relationship with elevation within the tidal frame generally produce more accurate results [18,19] and that modern site conditions provide the most appropriate analogue of those found in the fossil record. Here, we demonstrate that insufficient research has been undertaken on macrotidal regimes to establish the essential requirements of a ‘local’ training set for saltmarsh foraminifera in a macrotidal estuary.

In the Mersey Estuary, Wilson and Lamb [20] applied a UK-wide contemporary training set to establish a diatom-based tidal level transfer function for sea-level reconstruction, but found this to be seriously compromised by non-modern analogues. In this study, contemporary foraminiferal assemblages from Mersey Estuary saltmarshes were collected to:

(i) Establish whether their distributions exhibit vertical zonation controlled by elevation in a strongly macrotidal setting

(ii) Assess their suitability to form a modern training set with which to construct a tidal level transfer function and

(iii) Examine environmental and methodological considerations that impact on the efficacy of such a transfer function. Contemporary studies such as this are an essential precursor to any longer-term reconstructions that rely on the accuracy and precision of a sea-level transfer function.

Regional Setting

The Mersey Estuary is situated in NW England and is one of the largest estuaries in Britain. It can be divided into four separate physiographic regions: the Upper Estuary, Inner Estuary, Narrows and Outer Estuary (Figure 1a) [21]. The estuary experiences a macrotidal regime, with a spring tidal range of 8.4m and a mean tidal range of 4.5m at Liverpool [22]. Due to the constriction of the Narrows (Figure 1a) there is very limited penetration of waves from the Irish Sea and wave energy is relatively low in the Inner Estuary allowing the development of saltmarshes [21].

Figure 1: a: The Inner Estuary located with the Mersey Estuary NW England b: The location of the two investigated saltmarshes.

Study sites

Two saltmarshes were selected, based upon their location within the tidal frame, the foraminiferal preservation and species type (agglutinated or calcareous). Oglet Bay and Decoy Marsh are both located on the northern banks of the Inner Estuary (Figure 1b) [21]. Oglet Bay saltmarsh has a maximum length of 70m (cross-shore, i.e. front to back) and has an altitude between +4.36 to +5.87m OD, with a vertical range of 1.51 m. Vegetation zones are clearly defined across the saltmarsh with Phragmites spp. in areas at the back of the saltmarsh, followed by Scirpus maritimus, Spartina spp., Aster tripolium and Agrostis stolonifera in the lower parts of the saltmarsh. The upper elevation of the saltmarsh is restricted by a 12m cliff at the back of the saltmarsh, and the lower saltmarsh is truncated seaward by a 1m erosional cliff between the saltmarsh and mudflat. Decoy Marsh has a length of 200m, and is at an altitude ranging from approximately +5.29 to +5.88 m OD (vertical range of 0.60m). It is vegetated by low grasses (Festuca rubra) and horses currently graze part of the saltmarsh. Like Oglet Bay, Decoy Marsh is also characterised by an erosional cliff at the seaward saltmarsh edge.

Materials and Methods

Field sampling

A total of 105 surface samples were collected along five transects extending from the high to low saltmarshes (Figure 2). Three transects were sampled on Oglet Bay saltmarsh and two overlapping (spatially and vertically) transects on Decoy Marsh. Samples were collected during the winter months to early spring to minimize the influence of seasonal changes [23]. Due to the varied micro-topography of the saltmarsh it proved difficult to systematically collect surface sediment samples at vertical increments, hence samples were collected every 5-10 m horizontally [24,25] with the sample spacing decreasing (1-5m) as the saltmarsh surface gradient increased (i.e. higher spatial resolution on the high saltmarsh) (Figure 3). All sample elevations were obtained in m OD (Ordnance Datum - the GB reference mean sea level at Newlyn) to a levelling accuracy of c.2 mm using a Sokkia SET-5 Total Station and established Ordnance Survey benchmarks. Transect locations were chosen to capture a variety of sub-environments across the saltmarshes in relation to elevation, distance from main tidal access and differing vegetation cover.

Figure 2: Location of the transects from which modern foraminifera and environment data were collected from a) Oglet Bay and b) Decoy Marsh.

Figure 3: Environmental variables (elevation, salinity, organic matter, pH) from sampled transects. OBSS1, OBSS2 and OBSS3 are the three transects from Oglet Bay and DMSS1 and DMSS2 are the two transects from Decoy Marsh.

Sediment samples for analysis of both foraminifera and environmental variables were collected. A standardised volume of 10 cm³ was taken for foraminiferal analysis allowing comparisons with other studies including Scott and Medioli [26], Horton et al. [5], Horton et al. [27], Horton et al. [28], and Gehrels et al. [2,29]. Only the top 1 cm was collected as there has been found to be little effect of infaunal taxa over this depth in other UK saltmarshes [5,25]. Additionally, 30 cm³ of the same upper 1 cm of saltmarsh sediment was taken for analysis of other environmental variables, i.e. pH, salinity, grain size, organic matter [18,30,31].

Laboratory analysis

Foraminiferal analysis: Sample preparation followed an approach that is now well established amongst the research community who apply foraminiferal-based transfer functions to saltmarsh sediments for Holocene sea-level studies [6]. Sediments were stained using a Rose Bengal solution at the time of sampling [32] in order to identify specimens considered to be alive at the time of collection. 10 cm³ of sediment was then sieved through a 500 μm sieve and collected on a 63 μm sieve, following the methods of Scott and Medioli [26]. A wetsplitter was used [33] to divide the sample into 8 equal aliquots and the samples were then counted wet under a binocular microscope [24,25,33-35]. Sample aliquots were counted fully and repeated until at least 100 dead tests had been counted. Where species occurred in very low numbers, test counts were made of the full sample (>500) to ensure that the percentages and relative abundances remained the same in both. Species of Ammonia beccarii, Elphidium and Quinqueloculina were combined into generic groups. Quinqueloculina spp. includes oblonga and seminula, Elphidium spp. includes excavatum, incertum, and magellanicum, and Ammonia spp. includes variants of Ammonia beccarii [25,36].

Environmental variable analysis: Salinity and pH were measured using a 1:2 soil to water mix using a K&M 7002 conductivity and pH probe, which was calibrated for both pH and salinity. Conductivity was converted into salinity (ppt) using a standard equation (Salinity [ppt]=0.6679 [conductivity in mScm^-1] – 0.1513) following Gehrels and Newman [14]. Organic matter content was calculated using the Loss on Ignition (LOI) at 450°C for 4 hours [37]. Grain size was measured using a Coulter Laser Granulometer (LS200). The samples were digested before analysis using 20% hydrogen peroxide and heated gently on a hot plate. Less than 1cm³ of sediment from each digested sample was then mixed with Calgon to ensure disaggregation. The output data were processed using the computer program GRADISTAT [38]. The relative proportions of sand (0-4φ), silt (4-8φ) and clay (>8φ) in the minerogenic component (i.e. after organic digestion) were determined and are expressed as % by volume. In addition to grain size, pH, LOI, salinity and saltmarsh surface elevation, distance from the high tide shoreline at the back of the saltmarsh was used as a proxy for proximity to the main point of tidal ingress, i.e. the saltmarsh edge. Because the saltmarsh elevation along the sampling transects did not regularly increase with increasing distance from the seaward edge, the distance parameter was applied as a separate variable. As the transects differed in the vegetation present, particularly Phragmites spp., the presence and absence of Phragmites spp. was also included as a proxy variable to account for any significant influence from sub-surface freshwater and the potential influence of the reed stems on saltmarsh hydrodynamics, i.e. physical buffering of tidal inundation.

The linear transect sampling carried out in this study may suffer from the effects of spatial autocorrelation [39] where there is a tendency of sites close to each other to resemble one another more than randomly selected sites [40]. However, transect sampling remains the convention for collecting present-day samples [16,25] across the saltmarsh elevation gradient. The effects of spatial autocorrelation may be reduced in this study by combining several transects from the same saltmarsh, as well as combining data from the two saltmarshes [41]. Furthermore, transect number was included as a variable in the statistical analysis in order to determine whether transect location influences the species distribution significantly [42].

Statistical analysis

Statistical analysis was carried out using dead data only. The data were first converted into percentages and samples which had <50 dead foraminifera were removed. Most samples contained >100 individuals, apart from 5 samples which contained between 50-100 dead foraminifera. These samples were included in the dataset because the dominant species in these samples made up at least 50% of the total, with no co-dominant species and therefore are statistically reliable [43]. A total of 25 samples were removed due to low numbers, most of these samples were at high elevation within the tidal frame. The abundance of the remaining dead samples ranged from 28 to 3560 tests per 5 cm³, with an average of 1000 per 5 cm³. Species which contributed less than 5% were removed [44].

Initial inspection of the foraminifera species distribution across the different saltmarsh transects indicated that the dead species data from Decoy Marsh contained both autochthonous and allochthonous foraminifera. As the two Decoy Marsh transects overlapped in elevation and distance the observed difference in the contemporary datasets was interpreted as being due to the in-wash of allochthonous calcareous foraminifera from the outer estuary [45-47]. Indeed, macrotidal estuaries are particularly susceptible to the import of shelf foraminifera from the adjacent sea through suspended transport [48] or from storm reworking of sediments within the deeper parts of the estuary. Removing exotic species is important in the screening of modern data because the use of foraminifera as a proxy for sea level relies on the assumption that the dead assemblages are formed in-situ from living autochthonous populations. In this respect, the allochthonous input of foraminifera to assemblages are commonly removed for sea-level reconstructions [41,49,50].

The influence of allochthonous taxa on the Decoy Marsh dataset is illustrated in this paper by examining

(i) The agglutinated species only and

(ii) The ‘dead without exotic’ species.

Exotic species Brizalina spp., Cornuspira involvens, Fursenkoina spp, Fissurina spp., Globigerina spp., Glabratella milliettii, Bulimina spp. Eggerella spp. and Rosalina spp. were removed according to previously established criteria [51], which reduced the number of species to ten. The number of species then reduced to six when the agglutinated only data were included (Table 8).

Pearson’s Product correlation coefficients using PAST [52] were calculated for the environmental variables to determine the relationship between them. Tilia [53] was used to present the percentage foraminfera abundance data and to perform cluster analysis using CONISS [54]. Unconstrained cluster analysis was carried out on non-transformed data based upon unweighted Euclidean distance. Following Horton et al. [55], Pearson’s Product correlation coefficients were also calculated between cluster order and environmental variables to determine if the zonations are related to any of the environmental variables.

Detrended Correspondence Analysis (DCA) using Vegan [56] in R [57] was used to infer whether the data are unimodal or linear, and therefore which statistical methods are the most appropriate to use for that dataset [58]. The datasets have Standard Deviation (SD) gradients of between 2 and 3 SD and therefore either Canonical Correspondence Analysis (CCA) or Redundancy Analysis (RDA) ordination approaches were appropriate to use [59]. Both ordination methods were carried out and the method which produced a result in which the variables explain the most variance was chosen [60]. In this case, RDA was used for the constrained ordination analysis [61]. To test the significance of the environmental variables included in the ordination analysis, an automatic forward and backward stepwise model using permutation tests was adopted using Step in Vegan [56]. This builds the model so that it maximizes the adjusted r² at every step, and stops when the adjusted r² starts to decrease [62]. RDA was then re-run with the significant variables only (i.e. p<0.05).

Partial RDA (pRDA) was carried out to determine the effects of a single variable after removing the effects of the other variables [15,25,31,55]. The significance of individual variables was also calculated using permutation tests in Vegan [56].

A transfer function based upon the modern foraminiferal data was developed using C2 [63] to quantify the relationship between the taxa and elevation, and to assess the suitability of the data for a sealevel reconstruction. DCCA was carried out using elevation as the only variable [5,29,41] to determine whether unimodal or linear methods were appropriate for the transfer function method. Both datasets had a standard deviation of 2 for the elevation gradient. Therefore it is not clear whether the data are unimodal or linear. For this reason an ‘elevation’ transfer function, as a proxy for tidal inundation, was developed using the model Weighted Averaged Partial Least Squares (WAPLS). This method is used to assess the suitability of the data for a sea-level reconstruction. WAPLS produces different components in the results, each one increasing in complexity. The lowest component that gave acceptable results was chosen based upon the ‘principle of parsimony’ [64]. WAPLS has several advantages over other methods as it effectively considers the influence of additional environmental variables. WAPLS uses any structure present in the WA residuals that would otherwise be discarded by other methods [65] and has been found to be the method least affected by autocorrelation [66].

Performance was assessed using the statistical measures Root Mean Squared Error of Prediction (RMSEP) and the coefficient of determination (r²). These were estimated using the jack-knifing crossvalidation technique and are less biased than using the non-validated values, which can be underestimated and overestimated, respectively [64]. RMSEP_jack is a measure of the overall errors of the training set and is calculated to measure the prediction errors. r²_jack is calculated to measure the strength of the relationship between observed versus predicted values and is used to measure statistical predictive ability.

To refine the performance of the transfer function, outliers in the data may be identified and removed. However, it is essential to have robust criteria by which such ‘outliers’ are distinguished from accepted limits of variance in the data. Following Horton and Edwards [25] samples that have a poor fit to the regression line have a high residual distance from the elevation gradient. Therefore outliers are identified as those samples that have an absolute residual value (observed-predicted values) greater than that of the standard deviation of the elevation gradient [24,25,67,68]. These identified samples were removed, and the WAPLS model re-run with the new ‘pruned’ dataset. A total of 14 samples were removed from the total dead dataset, 16 samples were removed from the dead-exotics dataset and 10 samples were removed from the agglutinates only dataset.

Results and Interpretation

Environmental variables

Figure 3 shows the trend in the environmental variables along each of the five transects. It is stressed that the environmental data provide the winter season condition and, hence, we must be cautious in taking these data as the annual average. However, the overall approach is based on Horton and Edwards’ [23] findings that it is the winter season ecology that is most representative of the dead foraminiferal assemblages found in a saltmarsh core.

The environmental data (Figure 3) clearly illustrate the contrast in altitude and length (cross-shore) between the Oglet Bay and Decoy Marsh saltmarshes, with Decoy Marsh being higher in altitude and greater in length (from the high tide shoreline in the back-marsh to the saltmarsh edge). All transects and sites exhibit low salinities, ranging from 0 to 4.5%, thus indicating freshwater to marginally brackish water conditions as a consequence of the location of the sites at relatively high elevations within the inner estuary. All transects increase in salinity with increasing distance from the high saltmarsh, as expected due to the increasing tidal influence.

The organic matter content is similar for all transects and ranges from 66 to 6%: the lowest organic matter content can be seen in Decoy Marsh, compared with higher organic matter content in Oglet Bay (Figure 3). The very high organic contents are related to the dense vegetation on Oglet Bay high saltmarsh while the low values on Decoy Marsh are probably due to grazing by horses. The organic matter content decreases along all transects towards the seaward edge, which is again attributable to vegetation cover decreasing away from the high saltmarsh. The pH ranges from 8.4 to 7 and shows a decreasing trend from the back-marsh to the seaward edge that is common to all transects. The observed pH values are in reasonable agreement with the range of seawater pH. The spatial trend is interpreted as the result of freshwater runoff reducing the pH with decreasing altitude on the saltmarsh surface during the time period between successive high waters. The grain size for the transects is predominantly silt, making up approximately 80% of the composition, followed by clay which makes up roughly 10%, with sand contributing the least (<10%) (Figure 4). Transects OBSS2 and DMSS1 show no significant changes in grain size across the saltmarsh (front to back). In contrast, OBSS1 and OBSS3 show an increase in grain size in the high saltmarsh, with higher sand content up to 70%. From the saltmarsh stratigraphy [69] this sand component is of colluvial origin and derived from runoff and local erosion of the small cliff at the back of the saltmarsh - it is clearly not related to marine input. DMSS2 has an increase in grain size in the lower saltmarsh, with sand increasing to 40% of the composition and reduced silt content, which is more typical of grain size across a saltmarsh related to a tidal velocity gradient.

Figure 4: Grain size fractions of the minerogenic component for each transect. Relative proportions of sand (0-4φ), silt (4-8φ) and clay (>8φ) are expressed in % by volume. OBSS1, OBSS2 and OBSS3 are the three transects from Oglet Bay and DMSS1 and DMSS2 are the two transects from Decoy Marsh.

Foraminiferal distributions and clustering

Oglet Bay: Haplophragmoides wilberti occupies most of the Oglet Bay transects, particularly the high saltmarsh, with abundance ranging from 1% to 90%. Miliammina fusca is also a dominant species in all three transects with abundance ranging from 4% to 76%. In addition, Ammonia spp. dominates transect OBSS1, ranging from 1% to 90%, and Jadammina macrescens dominates OBSS3 where abundances range from 11% to 50% (Figure 5).

Figure 5: Relative percentages of dead foraminifera abundance for Oglet Bay, ordered by unconstrained cluster analysis based upon untransformed unweighted Euclidean distance. The cluster order is shown along with the transect which the sample was taken. OB1=OBSS1, OB2=OBSS2, OB3=OBSS3.

Figure 6 shows the results of the unconstrained cluster analysis based upon untransformed unweighted Euclidean distance for Oglet Bay, revealing four distinct zones: Three of the zones (I, II, IV) cover the same altitudinal range and species composition, including Haplophragmoides wilberti, M. fusca, with varying amounts of J. macrescens and Ammonia spp. (which was found in OBSS1 only, figure 5). Zone (III) has similar dominant species but with higher proportions of T. Inflata and calcareous species. Figure 6 shows the two main cluster groups and the altitudinal range they cover, illustrating a notable influence of elevation in the tidal frame on foraminiferal zonation.

Figure 6: Oglet Bay cluster zones defined by the first delineation from CONISS total sum of squares from the unconstrained cluster analysis based upon untransformed unweighted Euclidean distance, plotted against elevation. The dotted line showing the MHWS.

Decoy Marsh: The Decoy Marsh (Figure 7) assemblage can be divided into three zones. Zone I contains high proportions of M. fusca (43-53%) and B. pseudomacrescens (19-22%) with notable contributions from J. macrescens (6-24%) and Haplophragmoides wilberti (7- 24%). Zones II and III are characterized by greater abundances of J. macrescens, Elphidium spp., and Brizalina spp. with fewer M. fusca and Haplophragmoides wilberti. Zone II is then distinguished from zone III on the basis of high percentages of J. macrescens and low (and intermittent) contributions from Elphidium spp. and Brizalina spp.

Figure 7: Unconstrained cluster analysis based on unweighted Euclidean distance for relative percentages of dead foraminifera abundance on Decoy Marsh. The cluster order is shown along with the transect which the sample was taken. DM1=DMS1, DM2=DMSS2.

Figure 8 shows the results of the unconstrained cluster analysis based upon untransformed un-weighted Euclidean distance for Decoy Marsh, revealing three zones. Whilst there is considerable overlap in elevation between zones II and III, zone I is found at higher elevations within the tidal frame. Importantly, the entire elevational range of Decoy Marsh amounts to little more than 25 cm, so the notable higher position of zone I within the tidal frame illustrates some elevational control on foraminiferal distribution.

Figure 8: Decoy Marsh cluster zones defined by the first delineation from CONISS total sum of squares from the unconstrained cluster analysis based upon untransformed unweighted Euclidean distance, plotted against elevation. The dotted line showing MHWS.

Combined Oglet and Decoy Marsh: Oglet Bay and Decoy Marsh datasets are combined together in order to increase the sample size, and also to ensure that all sub-environments are effectively captured. The sample size increases to 82 samples, comparable with sample sizes of other studies ranging from 22 [70] to 165 (which includes 10 different sites) [55]. The unconstrained cluster analysis (Figure 9) reveals six different assemblage zones. At a general level, there are two overall groupings of the samples distinguished on the basis of the relative proportions of M. fusca and J. macrescens. Zones I-IV have high percentages of M. fusca (5-70%, generally >30%), high but variable percentages of Haplophragmoides wilberti (3-90%, generally >15%) and low percentages of J. macrescens (2-30%, generally <10%). Zone I is characterized by a high proportion of Ammonia spp. (15-90%, generally >30%), whilst zones II and III are distinguished on the basis of small percentages of calcareous taxa and T. inflata. For example, zone III has notable percentages of calcareous foraminifera (5-20%) and higher percentages of J. macrescens (8-26%) and T. inflata (3- 53%); comparatively, zone II has much lower frequencies of calcareous taxa (generally <5%), T. inflata (2% or less) and generally 5-25% B. pseudomacrescens. Zone IV shows the highest contribution from Haplophragmoides wilberti (55-90%). In contrast to zones I-IV, zones V and VI have low percentrages of M. fusca (0-20%, generally <12%) and high J. macrescens (5-74%, generally >25%). These two zones can be distinguished on the basis of zone V having high percentages of calcareous species, in particular Elphidium spp. (15-70%) and Brizalina spp. (4-16%), and zone VI having lower percentages of calcareous foraminifera (0-20%, generally <10%) but with the addition higher percentages (5-45%) of Haplophragmoides wilberti.

Figure 9: Relative percentages of dead foraminifera abundance for combined Oglet Bay and Decoy Marsh, ordered by unconstrained cluster analysis based upon untransformed unweighted Euclidean distance. The cluster order is shown along with the transect which the sample was taken. OB1=OBSS1, OB2=OBSS2, OB3=OBSS3, DM1=DMSS1, DM2-DMSS2.

The saltmarsh from which the samples were collected has a significant influence on the zonation. Zones I and IV relate almost exclusively to OBSS1, and zone III relates largely to OBSS3, with some contribution from OBSS1 and 2. In contrast, zones II, V and VI include contributions from both Decoy Marsh and Oglet Bay. The zonations generally overlap in altitude (Figure 10) although a decreasing altitudinal trend is apparent from zones II to III.

Figure 10: Combined Oglet Bay and Decoy Marsh cluster zones defined by the first delineation from CONISS total sum of squares from the unconstrained cluster analysis based upon untransformed unweighted Euclidean distance, plotted against elevation. The dotted line shows the MHWS.

Figure 10 shows that zone III occupies the lowest saltmarsh elevation of the combined dataset, this zone also contains a greater proportion of calcareous species (see above). Zone V also contains similar percentages of calcareous species (including greater numbers of Elphidium spp.). Unlike Zone III the altitudinal range of this zone is large due to the influence of calcareous species on Decoy Marsh (DMSS2) that is at a higher elevation. Zones I and II are similar in elevation range and are both characterized by the presence of Haplophragmoides wilberti and high percentages of M. fusca. However, zone I has much higher percentages of Ammonia spp. due to the presence of data from OBSS1 and OBSS2.

Despite the general overlap in elevation amongst the zones there is a clear separation between zones VI and III. These can be considered as the upper and lower saltmarsh zones, respectively. Similarly, zone I exhibits only slight overlap with zones VI and III and hence occupies the altitudinal range between the upper and lower saltmarsh environments. In contrast, zones IV, V and II exhibit progressively larger elevation ranges.

Clusters and environmental variables

Cluster order for Oglet Bay was correlated with the different environmental variables (Table 1). This reveals that the strongest correlations are both negative with elevation (r²= -0.65) and distance (r²= -0.53). In addition, although the respective r² values are low, organic matter and sand% also show a statistically significant negative correlation with cluster order, and silt% is positively correlated. The correlations between cluster order and environmental variables for Decoy Marsh (Table 2) show that elevation does not have a significant influence, with a p-value of 0.54 and an r² of -0.16. Cluster order is more strongly, and significantly, negatively correlated with distance, organic matter and silt%, and positively correlated with sand%. For the combined Oglet Bay and Decoy Marsh dataset, the correlation coefficients (Table 3) show no strong relationship with any environmental variable. However, both the presence of Phragmites spp. and elevation are both statistically significant negative correlations, having p-values of 0.001 and 0.045, respectively. Organic matter also exhibits a marginally significant negative correlation with cluster order. The relationship between elevation and cluster order is much weaker for the combined dataset than observed for Oglet Bay but is statistically significant.

Table 1: Correlation coefficients and p-values of cluster order from unconstrained cluster analysis based upon untransformed unweighted Euclidean distance and the measured environmental variables for Oglet Bay. R² values >0.5 and P-values <0.05 are in bold.

Table 2: Correlation coefficients and p-values of cluster order from unconstrained cluster analysis based upon untransformed unweighted Euclidean distance and the measured environmental variables for Decoy Marsh. R² values >0.5 and P-values <0.05 are in bold.

Table 3: Correlation coefficients and p-values of cluster order from unconstrained cluster analysis based upon untransformed unweighted Euclidean distance and the measured environmental variables for the combined Oglet Bay and Decoy Marsh dead foraminifera dataset. R² values >0.5 and P-values <0.05 are in bold.

Redundancy analysis

The results of the RDA for the Oglet Bay dataset (Figure 11a) show that the lower saltmarsh samples are related to high silt content, high clay content and high pH. The higher saltmarsh samples are more related to higher distance, higher elevation, higher organic matter and higher sand content. However there doesn’t appear to be a clear gradient. Table 4 illustrates that the 10 variables included in the RDA explain 55% of the total inertia. The RDA with only significant variables included, i.e. pRDA, can be seen in Table 5 in which distance, transect, pH, organic matter and elevation explain 52% of the total inertia. Here, most of the variance can be explained by transect number (14%), pH (13%) and inter-correlation (13%). The RDA for the Decoy Marsh data (Figure 11b and Tables 4 and 5) show that 73-77% of the variance in the dataset (depending on the number of variables used) can be explained by the environmental variables. Of these, only organic matter and clay% exhibit a clear explanation of the observed variance, with an important influence from intercorrelation.

Figure 11: RDA results for a) Oglet Bay, b) Decoy Marsh and c) the combined Oglet Bay and Decoy Marsh dead dataset. SD=sand, CL=clay, SI=silt, SL=salinity, PA=Phragmites spp., PH=pH, EL=elevation, DI=distance, TR=transect, OM=organic matter. AS=Ammonia beccarii spp., BS= Balticammina pseudomacrescens, BR=Brizalina spp., CY= Cornuspira involvens, ES= Elphidium spp., FS= Fursenkoina spp., GS=Glabratella milletti, HP= Haplophragmoides wilberti., HY= Haynesina spp., JM= Jadammina macrescens, MS= Miliammina fusca, QS= Quinqueloculina spp., TX=Textularia spp., TI= Trochammina inflata. a= OBSS1, b= OBSS2, c= OBSS3.

Table 4: RDA results for all datasets with all measured environmental variables.

Table 5: pRDA results for all datasets with only significant environmental variables included.

There does not seem to be any gradient or clear relationships identifiable from the RDA biplot of the combined dataset (Figure 11c), nor any obvious elevation gradient. Between 50 and 52% of the variance can be explained by the environmental variables (depending on the number of variables used), and 7 of the 9 significant environmental variables explain some level of variance (Tables 4 and 5). By far the greatest level of variance in the dataset is accounted for by intercorrelation between these environmental variables. Elevation explains only 2% of the total variance, but 4% of the explained variance (Table 5).

In order to address the potential confounding influences of allochthonous foraminifera, two further redundancy analyses were undertaken on

(i) combined ‘dead agglutinates only’ and

(ii) the combined dead datasets without exotic species.

The RDA ordination for the ‘agglutinated only’ data (Figure 12a) shows that no environmental variable influences the variance in the foraminifera more than any other. There is no clear gradient between variables.

Figure 12: RDA results for combined data a) agglutinates only b) exotics removed. SD=sand, CL=clay, SI=silt, SL=salinity, PA=Phragmites spp., PH=pH, EL=elevation, DI=distance, TR=transect, OM=organic matter. AS=Ammonia spp., BS= Balticammina pseudomacrescens, ES= Elphidium spp., HP= Haplophragmoides wilberti., HY= Haynesina spp., JM= Jadammina macrescens, MS= Miliammina fusca, QS= Quinqueloculina spp., TI= Trochammina inflata Tx=Textularia spp. a= OBSS1, b= OBSS2, c= OBSS3.

The RDA biplot of the combined dead data without the exotic species (Figure 12b) shows no strong allochthonous influence, with the species, sites and variables remaining the same as observed for the entire combined dataset. With all environmental variables included, the full combined dataset and the ‘combined minus exotics’ explain the same proportion of inertia as the entire combined data (Table 6).

Table 6: RDA results for selected elements of the combined Oglet Bay and Decoy Marsh datasets with all measured environmental variables.

In contrast, the ‘agglutinates only’ dataset explains less than half the variance. In terms of the variance explained by the different environmental variables, Table 7 shows that the results are almost identical between datasets with and without exotics, with elevation only explaining 2% of the total inertia in both. The RDA of the ‘agglutinates only’ dataset shows a lower proportion of the inertia being explained by the variables (37%), most of the variance being explained by the presence or absence of Phragmites spp. Elevation has no significant influence on the foraminifera distribution.

Table 7: pRDA results for combined Oglet Bay and Decoy Marsh with only significant environmental variables included.

Development of tidal level transfer function

A transfer function was developed using WAPLS and the combined Oglet Bay and Decoy Marsh data. The jack-knifed performance results from component 2 can be seen in Table 8. There was found to be a trend in the residuals in the WAPLS component 1 (i.e. WA). This trend reduced with the use of component 2 as WAPLS uses any structure present in the WA residuals. The full model with all samples included has an r²_jack of 0.52 and an RMSEP_jack of 0.24 m. When the data are pruned, removing 15 outlying samples as described previously, this improved the performance of the model, increasing the r²_jack to 0.56 and reducing the RMSEP_jack to 0.16 m.

Table 8: Results of component 2 of WAPLS analysis for the combined Oglet Bay and Decoy Marsh dataset using a dataset with all samples included and dataset which has been pruned. See methods for further details.

Transfer functions were also developed based upon the Oglet Bay data alone, the reduced datasets of ‘agglutinated species only’ and ‘without exotic species’. The performance measures from these can also be seen in Table 8. The use of the Oglet Bay data alone does little to improve the predictability and resolution of the ‘combined data’ transfer function, unless the statistical outliers are removed.

The ‘agglutinates only’ model has a much reduced number of species (6) and samples (67). Performance was found to be less than that of the full model with an r²_jack of 0.44 and RMSEP_jack of 0.24 m. However, when the data were pruned the model performance improved greatly to an r²_jack of 0.79 and RMSEP_jack of 0.13 m.

The model that performed least well used the dataset without exotic species. This model gave an r²_jack of 0.39 and RMSEP_jack 0.25 m, that only improved to the level of the ‘combined data’ when pruned.

Discussion

Controls on Zonation

Comparison of the Oglet Bay and Decoy Marsh zonation reveals a considerable difference in the elevational control. It appears that the height range across the Decoy Marsh surface is insufficient for elevation to have a significant control on zonation with the exception of zone I occupying a higher position than zones II and III. Hence, when the data are combined, the clear elevation differentiation observed in the zones on Oglet Bay becomes masked. This is clearly evidenced by the fact that the two lower saltmarsh foraminifera zones (zones III and I) are dominated by the Oglet Bay data whilst the remaining zones and the clear upper saltmarsh zone (zone VI) is a combination of Oglet Bay and Decoy Marsh data (Figure 10). In essence, the inclusion of the Decoy Marsh data contributes to the overlap between the observed foraminifera zonation. However it is still possible to identify distinctive lower saltmarsh (III) and upper saltmarsh (VI) foraminifera zones. Hence, considerable care must be taken to avoid over-emphasizing the high saltmarsh when sampling for the contemporary foraminiferal distribution and zonation, particularly when the upper saltmarsh zonation provides the upper bound to a tidal level transfer function.

Cluster order and environmental variables

An elevational control on the zonation of the Oglet Bay data is apparent but the influence of co-variance in the environmental variables across the saltmarsh weakens the significance of this control (Table 1). As elevation increases, so does the distance from the point of tidal ingress at the saltmarsh edge, and the organic content increases. Hence, elevation, distance and organic matter all exhibit statistically significant influences on cluster order, with correlation coefficients (r²) of -0.65074 (p-value 7.77E-09), -0.52832 (p-value 8.57E-06) and -0.35712 (p-value 0.004064) respectively. When we progress to the lesser range in elevation observed across Decoy Marsh, the influence of elevation on cluster order becomes negligible (r²=-0.1564 and p-value=0.535417), leaving distance and organic content as significant r² =-0.59755 (p-value=0.008824) and -0.40793 (p-value=0.092866 ) respectively (Table 2). In addition, grain size (which was statistically significant in the Oglet Bay data), becomes a more significant control on cluster order (Table 2). Hence, on Decoy Marsh, the elevation control is lost to the covariance of distance, organic matter and grain size across the saltmarsh surface. These emphasize the role of topography and distance to tidal ingress (saltmarsh edge) as an important control on saltmarsh foraminferal ecology. This is not unexpected when one considers the co-variance of saltmarsh vegetation, grain size and tidal flooding patterns on a low-amplitude saltmarsh surface [71].

Elevation returns as a control on cluster order for the combined dataset, along with organic matter (marginally) and vegetation (Table 3). We can infer from this that when the elevation range is limited, the inter-correlation of environmental variables across the saltmarsh surface becomes a more important control, i.e. the local microtopography of the saltmarsh surface has a more important influence on cluster order. Wilson and Lamb [20] report similar problems from the Mersey where significant local environmental diversity reduces the utility of regional diatom-based transfer function datasets. Likewise, Woodroffe and Long [19] noted poor predictive accuracy of combined datasets due to local variability in vegetation, organic content and tidal range between sites.

RDA

The pRDA data for Oglet Bay (Table 5) show that the environmental variables explain a significant proportion of the variance, and that elevation plays only a small role in this explanation. The amount of inertia explained by elevation for Oglet Bay is very low (4%) compared with previous studies where elevation explains between 3% to 32%, with an average percentage of 9% (Table 9). Distance from tidal influence contributes slightly more to the inertia in the data, i.e. 6%. Most of the variance in the species data was found to be related to the location of the samples on the saltmarsh, with transect number accounting for 14% of the variance. This is due to spatial differences in the environmental variables between the three Oglet Bay transects. It is important to consider that transect number incorporates all the differences between the sampling sites. Therefore any difference in pH, sand content and elevation between the transects will be incorporated in this one variable. The variables also correlate with each other (see Appendix Table A.1); transect number is strongly related to clay content (r²=0.76), silt content (r²=0.79), organic matter content (r²=0.68) and less strongly with distance (r²=0.47) and elevation (r²=0.43). This results in a large proportion of the variance being a combination of the variables (intercorrelation), contributing 13% of the total variance. When the transects are analysed individually [69], elevation has a much higher contribution towards the total inertia, with contributions ranging from 7-22%. In addition, the cluster order analysis reveals that elevation is important to the distribution of the foraminifera (r²=-0.65).

Table 9: Examples of different percentage contributions of elevation/SWLI (standard water level index), distance and/or flooding, to the total inertia in species distributions (foraminifera, diatoms, pollen) using pCCA or pRDA from several previous studies including the present study which includes significant variables only.

The amount of inertia in the combined data explained by the significant variables (sand, elevation, transect number, Phragmites spp. presence and salinity) remained similar (50%) and was still comparatively high, but the amount explained by elevation and/ or distance diminished from reasonable proportions explained by elevation for individual transects (i.e. 7-22%) falling to 2% when combined. This is similar to the studies by Charman et al. [42] and Roe et al. [72] who also showed that combining different datasets can reduce the amount of variance explained, and also the amount variance explained by elevation. One explanation is that the individual transect species data may have apparently stronger relationships with elevation and distance due to autocorrelation. This is related to the samples being taken along a transect, and therefore this relationship decreases when the data are added together [40]. The majority of the explained variance is made up of inter-correlations between the significant variables accounting for 32% of the total variance.

Importantly, whilst the pRDA suggests that the elevation control is not strong, the degree of variance in the foraminiferal data explained by the environmental variables for Oglet Bay is as good (if not better) than many published sea-level studies (Table 10), and the percentage explained by elevation is not significantly lower than comparable studies (Table 9). Equally importantly, these findings illustrate that cluster analysis has a propensity to over-emphasize the role of elevation on foraminiferal zonation, mainly due to the inter-correlation of other environmental variables with elevation.

Table 10: Examples of different transfer function performance measures (Root Mean Standard Error of Prediction, and R² using leave-one-out method of jack-knifing, unless otherwise stated) from modern surface data sets from around the world. Please refer to the online supplementary table for further details.

Transfer function

In addition to the statistical methods described above, the performance of a developed WAPLS transfer function [8,45,73] has been used to determine how well the modern dataset will perform a reconstruction of elevation (Table 8). The WAPLS pruned transfer function was found to have an r²_jack of 0.56 and an RMSEP_jack of c. 0.16 m (Table 8).

Acknowledging that some transfer function precision data are expressed as a percentage or as a Standard Water Level Index (SWLI), Table 10 illustrates that previous studies have returned RMSEP_jack, values which vary between 0.83 m [74] to 0.03 m [50] with most precisions being of the order of 0.1-0.2 m. Although there is no statistically significant pattern, the highest precision transfer functions are usually achieved for microtidal settings which have uncertainties below 0.1 m, or for studies that take a multiproxy approach, e.g. 0.05 m from Gehrels et al. [29] and 0.05 m from Kemp et al. [50]. Conversely, some of the lowest precisions are found in macrotidal environments (Table 10). With an RMSEP_jack of c. 0.17-0.18 m, and at best 0.13 m (see section 5.5), our findings offer a good level of precision for a macrotidal environment, although an r²_jack of 0.5 is at the lower end of the range of r²_jack values obtained from previous studies (Table 10, and also supplementary material).

Pruning assessment

From the pRDA and transfer function analyses (Table 8), it is clear that ‘pruning’ of the data can influence

(i) the proportion of variance explained by elevation and

(ii) the precision and predictability of the transfer function.

Whilst the dangers of pruning the data without due care and attention are acknowledged, we emphasize here that our analysis has been conducted to specifically examine the consequences of such data manipulation. Further, we stress that such data pruning is not proposed here for the first time: it has been applied routinely in many foraminiferabased tidal level transfer function studies [19,24,68,74]. Indeed, Gasse et al. [75] stress that it is accepted practice to remove samples with an absolute residual in the plot of observed against predicted elevation that is greater than one-quarter of the total range of the elevation gradient. Further, some sea-level studies remove samples from the lower part of the elevation range that exhibit nonlinearity with elevation, which may be caused by mixing of taxa by waves and other processes [76]. In several studies the lower altitudinal samples are removed as they are subjected to greater transport and bioturbation but also contain more in-washed foraminifera species [41,49,50]. Horton and Edwards [25] overcome the issue of allochthonous species by combining all in-washed (shelf) species into an ‘exotic’ species component. The overall consequence of reducing the entire dataset to pruned ‘agglutinates only’ data increases the r²_jack from 0.52 to 0.79 and the RMSEP_jack from 0.24 to 0.13 m. The precision and predictability both improve marginally but not substantially, confirming that the entire dataset can provide a reasonably accurate tidal level transfer function without further manipulation.

Although the ‘agglutinates only’ data should improve the elevational control on species distribution and zonation, it actually decreases the proportion of the variance explained by elevation. Indeed, it also decreases the overall explained variance relative to the combined data (Tables 5 and 7). This is because the analysis further limits the elevation range of the data (i.e. the agglutinated species are generally found on the upper saltmarsh). Hence, the reduced elevational range of the samples increases the relative ‘noise’ of the local saltmarsh surface environmental variable inter-correlations.

Removing the exotics did not improve the predictability and uncertainty in the ‘combined data’ transfer function (Table 8). This is probably due to the Elphidium spp that remain in the DMSS2 data. The total percentage of explained variance after removing the potential exotic species remains largely unaltered, and the percentage contributed by elevation remains low at 4% of the explained variance. More importantly, there is a concern here that this analysis excludes a proportion of the in-washed foraminifera that are obviously allochthonous but does nothing to remove any in-washed lower saltmarsh or upper saltmarsh species, which is certainly possible given the significant saltmarsh erosion observed in the Mersey. As this data processing does little to improve the extent to which the foraminifera distribution and zonation can be described as a function of elevation, the shortcomings of pruning seem to outweigh the benefits in this case. However, our results do show that appropriate and carefully evaluated pruning may increase transfer function accuracy and precision.

Elevation, Saltmarsh Micro-topography and Tidal Range

It is recognised that if there is insufficient elevational range in the sampling, the elevation control on the foraminiferal distribution and zonation is lost to other environmental variables that are intercorrelated and co-vary across the saltmarsh in accordance with its micro-topography and distance from tidal ingress [77]. Hence, the influence of elevation on the saltmarsh foraminifera is negligible on Decoy Marsh but significant at Oglet Bay. When combined (on the basis that a greater number of samples enhances the statistical robustness of the analysis) the elevation control is weakened. Hence, it would be better to not include the Decoy Marsh data and use only the Oglet Bay data, despite the r²_jack and RMSEP_jack of the transfer function not being substantially different (Table 8).

In developing a local ecological transfer function for sea-level reconstruction, it is stressed that sufficient elevational range must be achieved to limit the effects of microtopography and the flooding pattern of the saltmarsh surface. We have demonstrated that this is not achieved for Decoy Marsh but is just sufficient for Oglet Bay. This local sampling caveat has not really arisen to date, despite foraminiferal studies being undertaken on saltmarshes with macrotidal ranges of between 4.0 and 8.4 m [19,75,76,78-80]. However, both Horton et al. [55] and Gehrels et al. [29] found that the RMSEP of a transfer function tends to be lower for longer (greater elevational range) sampling transects. Similarly, Leorri and Cearreta [81] found a lower r²_jack when the extent of the elevation gradient was reduced.

The 25 cm elevation range of the Decoy Marsh samples in relation to the 8.4 m tidal range at Liverpool amounts to a c.3% proportion of the tidal range. If the same 25 cm sampling range is applied to meso- or microtidal environments then the proportion of the tidal range increases to 6-13% (4-2 m tidal range) and as much at 25% (1.0m tidal range). We propose here that the local sampling range must cover a sufficient proportion of the present-day tidal range to produce a viable tidal level transfer function, i.e. more than 10% of the spring tidal range (Oglet Bay sampling range c. 95cm=11% of the spring tidal range). If this is not achieved, the elevational influence on foraminiferal distribution and zonation will be lost in the noise of the saltmarsh topography/flooding pattern [20]. As a further consideration, whilst we have achieved the same elevational range of sampling in the combined Oglet Bay and Decoy Marsh dataset, there is a bias to the upper elevations (Decoy Marsh data) and hence the elevational control is weakened overall. One can infer from this that if the modern training set is skewed to the higher elevations, then the strength/reliability of a tidal elevation transfer function is likely to be compromised.

Summary and Conclusions

The aims of this study were to document the distribution of the foraminifera across two saltmarshes within a strongly macrotidal estuary, in order to determine the controls on foraminifera distribution and establish whether the dataset is appropriate to use for a local sea-level reconstruction. The species assemblages found are in good agreement with previous studies, with two main zonations across the saltmarsh: a high-to-middle saltmarsh zone occupied by Haplophragmoides wilberti, J. macrescens, and M. fusca, and a low saltmarsh zone composed of similar agglutinated species with increasing numbers of calcareous species including Brizalina spp., Elphidium spp., and Haynesina spp.

It is proposed that when the elevational range on the saltmarsh surface is low in comparison to the tidal range (i.e. <10%), the localscale inter-correlations of environmental variables across the saltmarsh surface become more significant than elevation in determining foraminiferal distribution and/or zonation. This is particularly challenging for macrotidal settings where the vertical range of the contemporary sampling is low compared with the tidal range.

pRDA shows that inter-correlation and within-site variability is too influential to place a robust statistical constraint on the influence of elevation on the saltmarsh foraminifera. However, when the pRDA is limited to the significant environmental variables alone, elevation accounts for 20% of the observed variance.

Elevational control is apparent in foraminiferal distribution and zonation, and in the correlation between environmental variables and foraminiferal clustering. The WAPLS transfer function shows that this control is sufficient to provide an elevation transfer function, as a proxy for tidal level, of limited resolution (RMSEP_jack 0.17 m) and moderate predictability (r²_jack=0.55). This is in reasonable agreement with previous tidal level transfer functions.

The predictability and precision of the elevation transfer function can be improved, as with other studies, by limiting the analysis to agglutinated species only and by pruning statistical outliers from the dataset (r²_jack=0.79, RMSEP_jack=0.13 m).

Although caution should be taken as some foraminifera appear to reflect changes in salinity, grain size and vegetation cover in addition to elevation and/or distance from tidal ingress, the modern distribution data from this study affirm the potential to reconstruct former sea-level from macrotidal saltmarshes, providing that appropriate sampling, data handling and statistical analysis are undertaken.

Acknowledgements

This research was partly funded by the Joanna Kinsey and the Farrington Hopkins trusts administered by the Department of Geography, University of Liverpool and the National Oceanography Centre in Liverpool. Timothy Shaw and Philip Woodworth are thanked for their assistance with fieldwork. Rob Smith the country park ranger from Halton Council is acknowledged for permission and access to Decoy Marsh, and Bill Warman - the slopes and deposit ground manager for the Manchester Ship Canal for access to Ince Banks. Irene Cooper and Alan Henderson are thanked for their help and assistance with laboratory work. Alan Bowden from the Liverpool World Museum is acknowledged for his help and advice. The authors wish to thank Rob Marrs for his help and guidance with the statistical analysis.

References

1. Donnelly JP, Cleary P, Newby P, Ettinger R (2004) Coupling instrumental and geological records of sea-level change: Evidence from southern New England of an increase in the rate of sea-level rise in the late 19th century. Geophys Res Lett 31.
2. Gehrels WR., Kirby JR., Prokoph A, Newnham RM, Achterberg EP, et al. (2005) Onset of recent rapid sea-level rise in the western Atlantic Ocean. Quaternary Sci Rev 24: 2083-2100.
3. Gehrels WR, Marshall WA, Gehrels MJ, Larsen G, Kirby JR, et al. (2006) Rapid sea-level rise in the North Atlantic Ocean since the first half of the nineteenth century. Holocene 16: 949-965.
4. Woodworth PL (1999) High waters at liverpool since 1768: the UK's longest sea level record. Geophys Res Lett 26: 1589-1592.
5. Horton BP, Edwards RJ, Lloyd JM (1999) A foraminiferal-based transfer function: Implications for sea-level studies. J Foramin Res 29: 117-129.
6. Gehrels WR (2000) Using foraminiferal transfer functions to produce high-resolution sea-level records from salt-marsh deposits, Maine, USA. Holocene 10: 367-376.
7. Hayward BW, Grenfell HR, Sabaa AT, Kay J (2010) Using foraminiferal faunas as proxies for low tide level in the estimation of Holocene tectonic subsidence, New Zealand. Mar Micropaleontol 76: 23-36.
8. Rossi V, Horton BP, Corbett DR, Leorri E, Perez Belmonte L, et al. (2011) The application of foraminifera to reconstruct the rate of 20th century sea level rise, Morbihan Golfe, Brittany, France. Quaternary Res 75: 24-35.
9. Kemp AC, Horton B, Donnelly JP, Mann ME, Vermeere M, et al. (2011) Climate related sea-level variations over the past two millennia. P Nat Acad Sci 108: 11017-11022.
10. Murray JW (1971) Living foraminiferainiferids of tidal marshes; a review. J Foramin Res 1: 153-161.
11. de Rijk S (1995) Salinity control on the distribution of salt-marsh foraminifera (great-marshes, Massachusetts). J Foramin Res 25: 156-166.
12. Patterson RT (1990) Intertidal benthic foraminifera biofacies on the Fraser River Delta, British Columbia. Micropaleontology 36: 229-244.
13. de Rijk S, Troelstra S (1999) The application of a foraminiferal actuo-facies model to salt-marsh cores. Palaeogeogr Palaeocl 149: 59-66.
14. Gehrels WR, Newman SWG (2004) Salt-marsh foraminifera in Ho Bugt, western Denmark, and their use as sea-level indicators. Dan J Geogr 104: 97-106.
15. Riveiros NV, Babalola AO, Boudreau REA, Patterson RT, Roe HM, et al. (2007) Modern distribution of salt marsh foraminifera and thecamoebians in the Seymour-Belize Inlet Complex, British Columbia, Canada. Mar Geol 242: 39-63.
16. Kemp AC, Horton BP, Culver SJ (2009) Distribution of modern salt-marsh foraminifera in the Albemarle-Pamlico estuarine system of North Carolina, USA: Implications for sea-level research. Mar Micropaleontol 72: 222-238.
17. Leorri E, Gehrels WR, Horton BP, Fatela F, Cearreta A (2010) Distribution of foraminifera in salt marshes along the Atlantic coast of SW Europe: Tools to reconstruct past sea-level variations. Quaternary Int 221: 104-115.
18. Horton BP, Edwards RJ (2005) The application of local and regional transfer functions to the reconstruction of Holocene sea levels, north Norfolk, England. Holocene 15: 216-228.
19. Woodroffe SA, Long AJ (2010) Reconstructing recent relative sea-level changes in West Greenland: Local diatom-based transfer functions are superior to regional models. Quaternary Int 221: 91-103.
20. Wilson GP, Lamb AL (2012) An assessment of the utility of regional diatom-based tidal-level transfer functions. J Quaternary Sci 27: 360-370.
21. Blott SJ, Pye K, van der Wal D, Neal A (2006) Long-term morphological change and its causes in the Mersey Estuary, NW England. Geomorphology 81: 185-206.
22. Admiralty (2010) Admiralty tide tables NP 201-07, Admiralty charts and publications, United Kingdom hydrographic office, Taunton.
23. Horton BP, Edwards RJ (2003) Seasonal distributions of foraminifera and their implications for sea-level studies. SEPM Special Publication 75: 21-30.
24. Edward RJ, Wright A, Van de Plassche O (2004) Surface distributions of salt-marsh foraminifera from Connecticut, USA: modem analogues for high-resolution sea level studies. Mar Micropaleontol 51: 1-21.
25. Horton BP, Edwards RJ (2006) Quantifying Holocene sea-level change using intertidal foraminifera: Lessons from the British Isles. Cushman Found Foramin Res Spec Publ 3-97.
26. Scott DB, Medioli FS (1980) Quantitative studies of marsh foraminiferal distributions in Nova-Scotia Canada implications for sea level studies. Cushman Found Foramin Res Spec Publ 1-58.
27. Horton BP, Edwards RJ, Lloyds JM (2000) Implications of a microfossil transfer function in Holocene sea-level studies. Holocene Land-Ocean Interaction and Environmental Change around the Western North Sea. Geological Society of London, Special publications, London, 41-54.
28. Horton BP, Whittaker JE, Thomson KH, Hardbattle MIJ, Kemp A, et al. (2005) The development of a modern foraminiferal data set for sea-level reconstructions, Wakatobi Marine National Park, Southeast Sulawesi, Indonesia. J Foramin Res 35: 1-14.
29. Gehrels WR, Roe HM, Charman DJ (2001) Foraminifera, testate amoebae and diatoms as sea-level indicators in UK saltmarshes: a quantitative multiproxy approach. J Quaternary Sci 16: 201-220.
30. Horton BP (1999) The distribution of contemporary intertidal foraminifera at Cowpen Marsh, Tees Estuary, UK: implications for studies of Holocene sea-level changes. Palaeogeogr Palaeocl 149: 127-149.
31. Horton BP, Culver SJ (2008) Modern intertidal foraminifera of the Outer Banks, North Carolina, USA, and their applicability for sea-level studies. J Coastal Res 24: 1110-1125.
32. Walton WR (1952) Techniques for recognition of living foraminifera. Contrib Cushman Found Foramin Res 3: 56-60.
33. Gehrels WR (1994) Determining relative sea-level change from salt-marsh foraminifera and plant zones on the coast of Maine, USA. J Coastal Res 10: 990-1009.
34. Scott DB, Hermelin JOR (1993) A device for precision splitting of micropaleontological samples in liquid suspension. J Paleontol 67: 151-154.
35. Gehrels WR (2002) Intertidal foraminifera as palaeoenvironmental indicators, in: Haslett, S.K. (Ed.), Quantitative Environmental Micropalaeontology. Arnold, London, 91-114.
36. Hayward BW, Holzmann M, Grenfell HR, Pawlowski J, Triggs CM (2004) Morphological distinction of molecular types in Ammonia - towards a taxonomic revision of the world's most commonly misidentified foraminifera. Mar Micropaleontol 50: 237-271.
37. Ball DF (1964) Loss on ignition as an estimate of organic matter ans organic carbon in non-calcareous soils. J Soil Sci 15: 84-92.
38. Blott S (2000) GRADISTAT version 4.0 ed, University of London, Surrey.
39. Legendre P (1993) Spatial autocorrelation - trouble or new paradigm. Ecology 74: 1659-1673.
40. Belyea LR (2007) Revealing the Emperor's new clothes: niche-based palaeoenvironmental reconstruction in the light of recent ecological theory. Holocene 17: 683-688.
41. Leorri E, Fatela F, Cearreta A, Moreno J, Antunes C, et al. (2011) Assessing the performance of a foraminifera-based transfer function to estimate sea-level changes in northern Portugal. Quaternary Res 75: 278-287.
42. Charman DJ, Roe HM, Gehrels WR (2002) Modern distribution of saltmarsh testate amoebae: regional variability of zonation and response to environmental variables. J Quaternary Sci 17: 387-409.
43. Hawkes AD, Horton BP, Nelson AR, Hill DF (2010) The application of intertidal foraminifera to reconstruct coastal subsidence during the giant Cascadia earthquake of AD 1700 in Oregon, USA. Quaternary Int 221: 116-140.
44. Fatela F, Taborda R (2002) Confidence limits of species proportions in microfossil assemblages. Mar Micropaleontol 45: 169-174.
45. Murray JW (1987) Biogenic indicators of suspended sediment transport in marginal marine environments - quantitative examples from SW Britain. J Geol Soc 144: 127-133.
46. Murray JW (2003) Patterns in the cumulative increase in species from foraminiferal time-series. Mar Micropaleontol 48: 1-21.
47. Horton BP, Murray JW (2006) Patterns in cumulative increase in live and dead species from foraminiferal time series of Cowpen Marsh, Tees Estuary, UK: Implications for sea-level studies. Marine Micropaleontol 58: 287-315.
48. Wang P, Murray JW (1983) The use of foraminifera as indicators of tidal effects in estuarine deposits. Mar Geol 51: 239-250.
49. Edwards RJ, Horton BP (2000) Reconstructing relative sea-level change using UK salt-marsh foraminifera. Mar Geol 169: 41-56.
50. Kemp AC, Horton BR, Corbett DR, Culver SJ, Edwards RJ, et al. (2009) The relative utility of foraminifera and diatoms for reconstructing late Holocene sea-level change in North Carolina, USA. Quaternary Res 71: 9-21.
51. Murray JW (2006) Ecology and applications of benthic foraminifera. Cambridge University Press, Cambridge.
52. Hammer Ø, Harper DAT, Ryan PD (2001) PAST: paleontological statistics software package for education and data analysis. Palaeontologia Electronica 4: 9.
53. Grimm EC (1991) Tilia 1.5.12, Illinois state museum.
54. Grimm EC (1987) CONISS: A FORTRAN 77 program for stratigraphically constrained cluster analysis by the method of incremental sum of squares. Comput Geosci 13: 13-35.
55. Horton BP, Edwards RJ, Lloyd JM (1999) UK intertidal foraminiferal distributions: implications for sea-level studies. Mar Micropaleontol 36: 205-223.
56. Oksanen J, Blanchet FG, Kindt R, Legendre P, O'Hara RB, et al. (2011) Vegan, ordination methods, diversity analysis and other functions for community and vegetation ecologists.
57. R. R Core Development Team (2009) R: A language and environment for statistical computing, Vienna, Austria: R Foundation for Statistical Computing.
58. Birks HJB (1995) Quantitative palaeoenvironemtal reconstructions, In: Maddy D. and Brew J.(eds), Statistical modelling of Quaternary science data, Technical guide No. 5., Quaternary Research Association Cambridge, 161-236.
59. Braak CJF ter, Prentice IC (1988) A Theory of Gradient Analysis. Adv Ecol Res 18: 271-317.
60. Puntí T, Rieradevall M, Prat N (2009) Environmental factors, spatial variation, and specific requirements of Chironomidae in Mediterranean reference streams. J N Am Benthol Soc 28: 247-265.
61. Leps J, Smilauer P (2003) Multivariate analysis of ecological data using CANOCO. Cambridge University Press, Cambridge.
62. Blanchet FG, Legendre P, Borcard D (2008) Forward selection of explanatory variables. Ecology 89: 2623-2632.
63. Juggins S (2007) c2, version 1.5.1. University of Newcastle.
64. Horton BP, Larcombe P, Woodroffe SA, Whittaker JE, Wright MR, et al. (2003) Contemporary foraminiferal distributions of a mangrove environment, Great Barrier Reef coastline, Australia: implications for sea-level reconstructions. Mar Geol 198: 225-243.
65. ter Braak CJF, Juggins S (1993) Weighted Averaging Partial Least-Squares Segression (WAPLS) – an improved method for reconstructing environmental variables from species assemblages', Hydrobiologia 269: 485-502.
66. Telford RJ, Birks HJB (2005) The secret assumption of transfer functions: problems with spatial autocorrelation in evaluating model performance. Quaternary Sci Rev 24: 2173-2179.
67. Jones VJ, Juggins S (1995) The construction of a diatom-based chlorophyll a transfer function and its application at three lakes on Signy Island (maritime Antarctic) subject to differing degrees of nutrient enrichment. Freshwater Biol 34: 433-445.
68. Engelhart SE, Horton BP, Roberts DH, Bryant CL, Corbett DR (2007) Mangrove pollen of Indonesia and its suitability as a sea-level indicator. Mar Geol 242: 65-81.
69. Mills HS (2011) Assessing the viability of using foraminifera from Mersey Estuary saltmarsh sediments to reconstruct former sea level. PhD thesis, University of Liverpool.
70. Sherrod BL (1999) Gradient analysis of diatom assemblages in a Puget Sound salt marsh: can such assemblages be used for quantitative paleoecological reconstructions. Palaeogeogr Palaeocl 149: 213-226.
71. Pratolongo P, Kirby JR, Plater AJ, Brinson MM (2009) Temperate Coastal Wetlands: Morphology, Sediment Processes, and Plant Communities. In: Perillo, G.M.E., Wolanksi, E., Cahoon, D.R. and Brinson, M.M. (eds.), Coastal Wetlands: An Integrated Ecosystem Approach. Elsevier, Amsterdam, .33-48.
72. Roe HM, Doherty CT, Patterson RT, Swindles GT (2009) Contemporary distributions of saltmarsh diatoms in the Seymour-Belize Inlet Complex, British Columbia, Canada: Implications for studies of sea-level change. Mar Micropaleontol 70: 134-150.
73. Massey AC, Gehrels WR, Charman DJ, White SV (2006) An intertidal foraminifera-based transfer function for reconstructing Holocene sea-level change in southwest England. J Foramin Res 36: 215-232.
74. Hill TCB, Woodland WA, Spencer CD, Marriott SB (2007) Holocene sea-level change in the Severn Estuary, southwest England: a diatom-based sea-level transfer function for macrotidal settings. Holocene 17: 639-648.
75. Leorri E, Horton BP, Cearreta A (2008) Development of a foraminifera-based transfer function in the Basque marshes, N. Spain: Implications for sea-level studies in the Bay of Biscay. Mar Geol 251: 60-74.
76. Gasse F, Juggins S, Khelifa LB (1995) Diatom-based transfer-functions for inferring past hydrochemical characteristics of African Lakes. Palaeogeogr Palaeocl 117: 31-54.
77. Hamilton S, Shennan I (2005) Late Holocene relative sea-level changes and the earthquake deformation cycle around upper Cook Inlet, Alaska. Quaternary Sci Rev 24: 1479-1498.
78. Allen JRL (1994) Morphodynamics of Holocene salt marshes: a review sketch from the Atlantic and Southern North Sea coasts of Europe. Quaternary Sci Rev 19: 1155-1231.
79. Edwards RJ Horton BP (2006) Developing detailed records of relative sea-level change using a foraminiferal transfer function: an example from North Norfolk, UK. Philos Trans A Math Phys Eng Sci 364: 973-991.
80. Horton BP, Corbett R, Culver SJ, Edwards RJ, Hillier C (2006) Modern saltmarsh diatom distributions of the Outer Banks, North Carolina, and the development of a transfer function for high resolution reconstructions of sea level. Estuar Coast Shelf S 69: 381-394.
81. Leorri E, Cearreta A (2009) Recent sea-level changes in the southern Bay of Biscay: transfer function reconstructions from salt-marshes compared with instrumental data. Scientia Marina 73: 287-296.