Compositional Data Analysis (CoDA) as a tool to study the (paleo)ecology of coccolithophores from coastal-neritic settings off central Portugal

Highlights

• Recent coccoliths as (paleo)ecological proxies along the coastal–oceanic transition

• Classical vs. compositional analytical methods (i.e., CoDA)

• Advantages of using the isometric log-ratio approach

• Submarine canyons vs. adjacent shelf-slope regions, west of Portugal

Abstract

Whereas using the species percentages is the standard analytical procedure used to infer species ecological preferences, independently of taphonomical effects, the closure problem associated with closed number systems and subsequent inconsistency of determining percentages may lead to spurious correlations, biased statistical analysis and misleading interpretations. To avoid these problems, we applied Compositional Data Analysis (CoDA) to investigate the (paleo)ecological preferences and spatial distribution of coccolith assemblages preserved in seafloor sediments, using as a case-study the central Portuguese submarine canyons and adjacent shelf-slope areas. Results from using the isometric log-ratio (ilr) approach from CoDA are compared with results from using classical analytical methods, and further discussed.

While providing scale invariance and subcompositional coherence, CoDA is revealed to be a consistent statistical tool to infer the (paleo)ecological preferences of coccolithophores, corroborating earlier work based on from percentage determinations. Results of this study clearly confirmed the coastal-neritic distribution of coccoliths from Gephyrocapsa oceanica, Helicosphaera carteri and Coronosphaera mediterranea, whereas coccoliths from Calcidiscus leptoporus, Umbilicosphaera sibogae, Umbellosphaera irregularis and Rhabdosphaera spp. more typically occur offshore. Differences between canyons and adjacent shelf and slope areas were also confirmed, namely the (a) greater importance of the coastal-neritic assemblage possibly resulting from local and persistent nutrient pumping in these areas, and (b) stronger mixing of coccoliths from coastal and oceanic species in upper canyon reaches, resulting from the focused coastward advection of more oceanic water masses along their axes.

Unlike the results from both ilr-coordinates and the percentage approaches, both coccolith concentrations and fluxes showed that spatial trends in which the species ecological inter-relationships appear to be masked by taphonomical phenomena, especially towards the coast and in the canyons, suggesting the two latter approaches are not suitable to perform (paleo)ecological inferences in more dynamic coastal-neritic settings.

Our study suggests that the (paleo)ecological signal preserved in the studied sediment samples is persistent enough to be revealed by both CoDA and percentages. Yet, given that CoDA provides the only statistical solution to coherently draw (paleo)ecological interpretations from compositional data, our recommendation is that CoDA should always be used to test and validate any ecological signals obtained from percentage distributions.

Introduction

Coccolithophores are the predominant phytoplankton group within calcareous nannoplankton, commonly used as (paleo)environmental proxies and markers of oceanographic processes (e.g., Ziveri et al., 2004, Silva et al., 2008, Guerreiro et al., 2013) due to their exceptional fossil record in both open ocean (Ziveri et al., 2004) and continental shelf/slope sediments (Cachão and Moita, 2000, Guerreiro et al., 2005, Guerreiro et al., 2015). Although affected by various post-mortem (biostratonomical) mechanisms, studies indicate that coccolith thanatocoenoses preserved in surface sediments can be closely related to the coccolithophore communities dwelling in the overlaying photic layer (e.g., Abrantes and Moita, 1999, Baumann et al., 2000, Kinkel et al., 2000, Sprengel et al., 2002, Boeckel and Baumann, 2008, Guerreiro et al., 2015) due to rapid transfer mechanisms mediated by zooplankton grazing (Steinmetz, 1994, Balch, 2004). Nevertheless, the correspondence between the coccolithophore living communities and the coccolith species assemblages preserved in the seabed is complex, particularly those close to the coast and in the context of a continental margin dissected by submarine canyons, where both continental and marine factors interplay (see Roth, 1994, Steinmetz, 1994).

Productivity of coccolithophores is primarily a function of light and nutrients, modulated by hydrological parameters like temperature, salinity, turbulence and turbidity (see Margalef, 1978). Seasonal variability in controlling parameters results in successions of seasonally distinct coccolithophore biocoenoses. The coccolith fossil record preserved in the seafloor sediment results from this primary signal, which is further affected by several taphonomical phenomena (i.e., necrolysis, biostratonomy, variations in sediment accumulation rate, bioturbation and diagenesis) that act upon the coccospheres/coccoliths as illustrated in Fig. 1. In other words, whereas a certain productivity represents the cumulative coccolithophore production of a certain region of the ocean, the fossil record is the net cumulative average over 10^− 1–10⁴ years of continuous productivity, more or less distorted by taphonomy.

In addition to the primary ecological signal, the coccolith assemblages preserved in seafloor sediments will also contain allochthonous coccolithophores (i.e., advected from adjacent water masses), as well as reworked fossil or subfossil specimens resuspended from the bottom of continental shelf and slope regions. Such contribution is particularly important where continental margins are dissected by submarine canyons, given that these are regions of enhanced water column and sedimentary dynamics (i.e., intensified vertical water motion, internal tides, gravity flows; De Stigter et al., 2007, De Stigter et al., 2011, Oliveira et al., 2007, Arzola et al., 2008).

Once deposited at the seafloor, the coccolith thanatocoenosis will be subjected to the effects of dilution with other sedimentary material (including reworked coccoliths), and to mixing by organisms living in the benthic sediment layer, the latter of which can result in more or less homogenised coccolith taphocoenoses. After burial, the association will be further subjected to diagenesis occurring on longer time scales (i.e., 10²–10⁶ years). Processes such as dissolution and recrystallization will tend to preferentially eliminate the smaller and more delicate forms while favouring the preservation of the larger and more robust ones, resulting in an orictocoenosis only partially representing the original assemblage (Fig. 1).

To extract the (paleo)ecological signal from the fossil record, in order to make ecological and subsequent oceanographic inferences, the effects of advection and reworking, variations in sedimentation and diagenesis must be quantified and removed.

Distortions induced by diagenesis can be significantly reduced by: (1) studying late Holocene sediment records for which diagenesis can be assumed to be negligible, and/or (2) selecting species with a similar degree of resistance to dissolution/recrystallization. The removal of such effects results in taphocoenoses that are now only affected by biostratonomical advection, sediment reworking and variations on sediment accumulation rate. To remove the latter effects in order to access the thanatocoenosis' (paleo)ecological signal, determining the coccolith (or nannolith s.l.) fluxes (nanno/cm²/year) from a constant sedimentation rate may be a solution. However, current radiometric methods commonly used for determining sediment accumulation rates in continental margin settings provide only averages over decadal to centennial time periods, thus lack the desired finer resolution. In addition, effects of bioturbation can often not be accounted for in the calculation of sediment accumulation rate, leading to considerable errors especially in slowly-accumulating sediments from open continental slopes (see Carpenter et al., 1982, Fuller et al., 1999, Boer et al., 2006). Finally, in regions of high sediment accumulation rates such as submarine canyons (De Stigter et al., 2007, De Stigter et al., 2011) coccolith fluxes derived from present day productivity are likely to end up mixed with resuspended/reworked coccoliths, due to the fact that these major submarine valleys often act as morphological depocentres of fine particles in transit along the shelf.

Determining species percentages relative to the sum of total (or selected) common coccoliths are a standard analytical procedure to eliminate the effect of the (common) taphonomical factors that indiscriminately act upon the biocoenoses, thus allowing to infer the species ecological inter-relationships independent of the effects from sediment dilution and/or bottom dynamics. Such factors are, thus, assumed to affect all coccolithophore species of similar size (≥ 3 μm) equally. Yet, given that coccolith assemblages fall into the category of compositional data, they are affected by spurious correlations and subcompositional incoherence, for which percentage determinations often lead to partial loss and/or misunderstanding of part of the information within compositional natural systems (Pawlowsky-Glahn and Egozcue, 2001, Buccianti, 2013). Even if one applies the strategies mentioned above, the interplay of all these environmental factors is likely to induce subcompositional incoherence if non-log ratio methods are used to investigate compositional data.

Compositional data can be defined as any data in which the components represent “parts of a whole”. They carry only relative information between parts, and are usually represented as percentages, in which case, their sum corresponds to a constant, typically 100% (Parent et al., 2012). Because of this, compositional data are an easy target of spurious correlations and subcompositional incoherence that typically affect any data measuring parts of the same whole. An example: considering an ecological subsystem composed of a few selected species (sp₁, sp₂, sp₃, … sp_n), the behaviour of any two components will randomly change their signs (i.e., if they are positively or negatively correlated) and meanings depending on the other components involved (Pearson, 1897, Tolosana-Delgado, 2012, Buccianti, 2013). Ideally, if the above mentioned strategies (1) and (2) have been applied, results should not depend on whether we take a few, several, or all of our components or species, neither on whether we scale them up and down by arbitrary numbers, nor even whether we choose our samples to sum up to 100% or leave them untouched (Aitchison, 1997). But due to the properties of compositional data, the results obtained from their percentages may in fact vary significantly.

To avoid these problems, and to ensure that all the available information is being extracted from the data, two fundamental properties should be inherent to the methods applied when studying compositional data: “scale invariance” and “subcompositional coherence”. This means that whether one represents the data in percent, ppm or some other unit, and whether one closes the analysis for two, three or ten species, the results, i.e., the relationships between the species, will be the same and coherent among themselves.

Compositional Data Analysis (CoDA), based on the idea that we can only assess relative changes between components, provides a set of statistical tools and geometric concepts built with the purpose of allowing results to satisfy scale invariance and subcompositional coherence (Aitchison, 1997). By avoiding spurious correlations and negative bias, the log-ratio approach allows obtaining an understanding on how natural phenomena work while extracting all the information contained in data variability (Buccianti, 2013). Although similar results may be achieved from both classical and compositional approaches, depending on which are the typical values of the percentages and their spread, it is not possible up to now to know the weight of this effect a priori, or even to measure it (see Pawlowsky-Glahn and Egozcue, 2001, Tolosana-Delgado, 2012).

In our study, CoDA is used to study the (paleo)ecology of coccolithophores from the central Portuguese margin. The isometric log-ratio (ilr) approach (e.g., Egozcue et al., 2003, Egozcue and Pawlowsky-Glahn, 2005, Thió-Henestrosa et al., 2008, Tolosana-Delgado, 2012) was applied to (1) validate (paleo)ecological trends previously inferred from the species percentages (Guerreiro et al., 2015), (2) confirm the potential of coccoliths as proxies of environmental phenomena intensified in submarine canyons, and (3) introduce CoDA as a method to validate and obtain consistent (paleo)ecological interpretations. Results obtained from CoDA are compared with results from classical approaches (i.e., coccolith concentrations, species relative percentages and coccolith fluxes) and are further discussed.