Benthic species of the Kerguelen Plateau show contrasting distribution shifts in response to environmental changes

Abstract Marine life of the Southern Ocean has been facing environmental changes and the direct impact of human activities during the past decades. Benthic communities have particularly been affected by such changes although we only slowly understand the effect of environmental changes on species physiology, biogeography, and distribution. Species distribution models (SDM) can help explore species geographic responses to main environmental changes. In this work, we modeled the distribution of four echinoid species with contrasting ecological niches. Models developed for [2005–2012] were projected to different time periods, and the magnitude of distribution range shifts was assessed for recent‐past conditions [1955–1974] and for the future, under scenario RCP 8.5 for [2050–2099]. Our results suggest that species distribution shifts are expected to be more important in a near future compared to the past. The geographic response of species may vary between poleward shift, latitudinal reduction, and local extinction. Species with broad ecological niches and not limited by biogeographic barriers would be the least affected by environmental changes, in contrast to endemic species, restricted to coastal areas, which are predicted to be more sensitive.


| INTRODUC TI ON
The consequences of global climate change in the polar seas are predicted to lead to warmer, fresher, and more acidic waters, in addition to more extreme climatic events and seasonal variations than actual conditions (Allan et al., 2013;Gutt et al., 2015). Significant changes have already been recorded in Antarctic waters; for instance, sea surface water temperature in the western Antarctic Peninsula has increased by 1°C over the last half-century (Turner et al., 2013). All these changes are critical for Antarctic organisms as they lead to a decrease in habitat suitability for the species (Clarke et al., 2007;Doney et al., 2011). Sub-Antarctic ecosystems are confronted with the direct and indirect impacts of climate change too (i.e., glacier retreat, temperature increase, decrease in precipitations), with the combined effects of these multiple stressors also leading to the prevalence of favorable climatic conditions for introduced species and, consequently, to alterations in the pristine marine life (Allan et al., 2013;Byrne, Gall, Wolfe, & Agüera, 2016;Kargel et al., 2014;Molinos et al., 2015;Pendlebury & Barnes-Keoghan, 2007;Smith, 2002).
For instance, cidaroids were listed as vulnerable marine ecosystem (VME) indicator taxa by CCAMLR (Convention for the Conservation of Antarctic Marine Living Resources, CVD code) because of their rich epibiont assemblages (Hardy, David, Rigaud, De Ridder, & Saucède, 2011;Linse et al., 2008), brooding behavior, lack of motility, and sensitivity to fishing activities (CCAMLR 2009).

| Studied area
The Kerguelen Plateau is a vast and remote area of the Southern Ocean that displays unique oceanographic features and proximity between marine fronts generating strong latitudinal temperature and salinity gradients (Moore & Abbott, 2002;Park et al., 2014). Important micronutrient releases, including iron, and high chlorophyll a concentrations are present on the eastern margin of the plateau. This contrasts with the "High Nutrient Low Chlorophyll" waters reported in most of the Southern Ocean (Koubbi et al., 2016;Park et al., 2014). High diversity levels in pelagic and benthic ecosystems are also described on the Kerguelen Plateau in comparison with the surrounding oceanic areas (Féral et al., 2016;Koubbi et al., 2016). The Kerguelen Plateau makes part of the French and Australian protected areas and aggregates substantial conservation issues for marine biodiversity due to fast environmental changes and the impact of fisheries activities (CCAMLR 2008, Koubbi et al., 2016Welsford, Constable, & Nowara, 2011;Welsford, Ewing, Constable, Hibberd, & Kilpatrick, 2014 (Allan et al., 2013). This makes marine biodiversity of the region particularly at risk with regard to environmental changes. Alteration in marine biodiversity and ecosystem functioning is particularly expected to impact coastal marine areas of the Kerguelen Islands (CCAMLR 2008, 2013, Hureau, 2011
The dataset contains presence-only data of 201 echinoid species collected in the Southern Ocean from the Antarctic coasts to 45°S latitude. This dataset is a compilation of data collected during oceanographic campaigns undertaken between 1872 and 2015. Four species with contrasting ecological requirements were selected in the dataset for this study. The four selected species are known by a sufficient number of presence records to perform robust species distribution models. These four species are common on the Kerguelen Plateau and constitute substantial representatives of Antarctic benthic ecosystems (De Ridder, David, & Larrain, 1992;Díaz, Féral, David, Saucède, & Poulin, 2011;Hardy et al., 2011;Linse et al., 2008;Moya, Saucède, & Manjón-Cabeza, 2012;Poulin & Féral, 1995). Namely, we selected Abatus cordatus and Brisaster antarcticus, two species endemic to sub-Antarctic regions, and Ctenocidaris nutrix and Sterechinus diadema that present broader distribution ranges in the Southern Ocean ( Figure 1).
Duplicate records that fell on one single grid-cell pixel of environment layers were removed from the dataset.

| Environmental data
The environmental descriptors used in this study were generated and described by Fabri-Ruiz et al. (2017). They are available as raster layers collected from different sources and modified to fulfill modeling requirements at the scale of the Southern Ocean.
Environmental data cover the extent of the Southern Ocean (<45°S) at a grid-cell resolution of 0.1 degrees (around 10 km). Neighborjoining interpolation was applied to correct for missing values that may interfere with certain distribution modeling approaches. The dataset contains environmental descriptors for the six decades included between 1955 and 2012. Environmental data were processed using the functions proposed by the R (R Core Team 2015) SDMPlay (Guillaumot, Martin, Eléaume, & Saucède, 2016) and raster packages (Hijmans 2016).
Predictor selection is a major concern that can alter modeling performances (Braunisch et al., 2013;Petitpierre, Broennimann, Kueffer, Daehler, & Guisan, 2017). Here, we tested for collinearity between predictors and removed one descriptor from the initial dataset for VIF > 5 (variance inflation factor with the stepwise F I G U R E 1 Distribution of presence-only records available for the four studied species (occurrence duplicates were removed)
Oxygen concentration and POC export layers were not available for future projections. Similarly, chlorophyll a, oxygen concentration, POC export, and ice coverage layers were not available for past projections. Therefore, these environmental parameters were considered constant and similar to the present values (Table S1).
Presence-only methods imply using background data to be selected in the study area. In the Southern Ocean, sampling effort is spatially contrasted and such spatial heterogeneities can alter modeling performances (Araújo & Guisan, 2006). We corrected for spatial bias using a target-background sampling method (Phillips et al., 2009). The kernel density estimation (KDE) of visited pixels was estimated (i.e., grid cells in which at least one echinoid is recorded, according to the database compiled by Fabri-Ruiz et al. (2017)). KDE layer is a proxy for the survey effort that is used to spatially weight background sampling and compensates for the weight of frequently visited sites in the models. As suggested by Barbet-Massin et al. (2012), we selected a number of background records similar to the number of presence-only records available and applied a 100-fold replication of the background sampling in each case.
To take into account the limitations of model extrapolation, modeling areas were limited in geography and depth. On the basis of species records and ecology (David et al., 2005), we considered 1,000 m depth as a maximal extrapolation threshold for A. cordatus and S. diadema, and 1,500 m depth for B. antarcticus and C. nutrix. Boundaries in latitude and longitude were species-specific and determined according to the most extreme positions of species records.

| Species modeled ecological niches
The modeled environmental spaces occupied by the four stud- S. diadema is high (−1°C to +5°C) with preferences for low temperatures on the extent of the Southern Ocean ( Figure S2).

| Environmental shifts and evolution of ecological niche space
To compare the size of species realized niches between the three periods ( Figure 6), the environmental subspace occupied by each species was delineated by plotting the values of environmental descriptors that contribute the most to the models (Figure 2a'-d').

| SDM limitations and perspectives
Modeling species distribution in the Southern Ocean is challenging.
The paucity of available data is a major limitation to analyses that are restricted to presence-only data models, usually considered less reliable and less efficient than presence-absence or abundance data models (Brotons et al., 2004 of sampling in the present case study), which can influence modeling performances (Guillaumot et al., in press;Newbold, 2010;Tessarolo et al., 2014). SDM performed with spatially biased presence-only data must consider these limitations and apply appropriate algorithms, protocols, and corrections (Barbet-Massin et al., 2012;Guillaumot et al., in press;Phillips et al., 2009;Proosdij et al., 2016).
Strengthening sampling effort to characterize and model the entire distribution area of widely distributed species is one of the main priorities of Antarctic science, as recently pointed out by Kennicutt et al. (2014). Improving our knowledge of the Southern Ocean marine life also includes the development of efficient and robust modeling approaches. For instance, other environmental descriptors should be added to models in order to better characterize and depict species niches (Austin & Van Niel, 2011;Bradie & Leung, 2016).

AUC values differ between SDM, with high and stable values for
A. cordatus and lower scores for B. antarcticus and S. diadema. In previous works, Qiao, Soberón, and Peterson (2015) discussed the link between species niche width and the evaluation of SDM performances using true skill statistics and kappa indices. SDM produced for narrowniche species were proved to be associated with high sensitivity and high specificity scores. This is in line with the present results that show the highest AUC values for models performed for the narrowniche species, A. cordatus, and the lowest values for the wide-niche species, S. diadema. When using AUC to evaluate the performance of models performed with presence-only data, the maximum AUC value is given by the formula AUC = 1 − α/2, in which α corresponds to the fraction of the study area occupied by the species (Phillips, Anderson, & Schapire, 2006;Proosdij et al., 2016;Raes & ter Steege, 2007).
Considering the extent of the species respective distribution areas, A. cordatus is the species with the smallest fraction of area coverage (α), which can account for the high AUC values of the model. TA B L E 2 Comparison of model outputs using the metrics developed by Crase et al. (2015); see equation details in Table 1. Values correspond to mean and standard deviation of 100 model replicates. Suitable area defined as the number of pixels for which species distribution probabilities are higher than the MaxSSS threshold value

| Major environmental drivers
Environmental descriptors that contribute the most to SDM vary between species according to the different ecological niches. For instance, the distribution of the deposit-feeder B. antarcticus is strongly correlated with chlorophyll a concentrations and the species distribution is mainly predicted in regions with chlorophyll a blooms such as in the northeast of the Kerguelen Islands, in the vicinity of the Polar Front, and near the coasts of Crozet Islands (Park et al., 2014). In contrast, the nearshore species A. cordatus is mainly correlated with the values of seafloor temperature amplitudes, which are the highest in the shallow waters of the Kerguelen Islands.
The present results show the importance of using amplitude values in SDM, in association with other parameter metrics. They contribute to the SDM performed for the four echinoid species of this study as major descriptors. Bradie and Leung (2016) already discussed the importance of including descriptors of seasonal means and extremes in models. These descriptors were proved to further account for species distribution patterns than annual means, considering their stronger relationship with species niche width and ecological traits (i.e., growth and survival; see Franklin, 2009).

| Species responses to environmental change
In the present work, we could generate robust models to assess the influence of changing environmental conditions on species distribution range, both in the geography and in the environment.

| Effects of environmental change and conservation strategies
Water seafloor temperature and salinity were shown to have significantly varied since the 1950s, but major changes are still to come in a near future according to IPCC scenario RCP 8.5 (IPCC Fifth Report, 2013). Modifications in the seasonal amplitude of temperature and salinity variations should have decisive effects on costal marine ecosystems (CCAMLR 2008(CCAMLR , 2013Féral, Beurier et al., 2016;Gutt et al., 2018;Hureau, 2011;Sahade et al., 2015;Schram et al. 2016;Smale & Barnes, 2008).

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
CG, AM, ME, and TS conceived the ideas and designed methodology; SFR provided a part of the data; and CG and TS wrote the manuscript and all the remaining authors contributed critically to the drafts and gave final approval for publication.