**1. Introduction**

Transitional waters, including estuaries and coastal lagoons, are highly heterogeneous ecosystems, being characterised by the presence of strong gradients in water and sediment properties and composed of a diverse mosaic of morphologies and biogenic structures [1,2]. This makes them highly valuable ecosystems supporting unique biological communities. Nekton fauna (fish and swimming invertebrates) play a central role in transitional waters, mediating multiple ecological processes and including species of commercial and conservation interest [1,3–5]. The distribution of biological communities in transitional waters is driven by multiple environmental factors, among which salinity is crucial in determining the organism responses at the physiological level [3,6]. The nekton community structure, in particular, may be a ffected by the di fferent species tolerances and preferences to salinity, with migratory, marine and freshwater straggler taxa being especially influenced by spatial and inter-annual variations in salinity levels within estuarine ecosystems [3,7–11].

A variety of anthropogenic pressures a ffect transitional water ecosystems, which may lead to habitat degradation, alterations of ecological processes and depletion of biological communities, ultimately impairing the ecosystem status and functionality [12,13]. The restoration of both abiotic

and biotic components of transitional water ecosystems is recognised as a strategic approach to both enhance the ecological status of degraded transitional water ecosystems (sensu Dir. 2000/60/EC Water Framework Directive, WFD) and tackle the loss of biodiversity [14–16]. In recent years, a variety of schemes have been carried out in estuaries and coastal lagoons all over the world, aiming at re-establishing specific habitats previously degraded or lost [17,18] and enhancing the status of faunal communities [19–22]. Due to the central role of salinity in transitional water ecosystems, its managemen<sup>t</sup> by means of hydrological and morphological improvement have been of particular interest for scientists and practitioners in these environments [16,23,24], and in many instances, the control of salinity has been proposed as a measure to achieve ecological restoration. For example, increasing and stabilising salinity levels in enclosed lagoon basins by enhancing connections with the sea would improve trophic status and conservation of vulnerable habitats such as saltmarshes and seagrass meadows [25], as well as fishery yields and faunal diversity [26]. On the other hand, recreating the salinity gradient in estuaries and lagoons subjected to previous river diversions and the unsustainable use of freshwater could significantly contribute to the restoration of key ecological components [27], as well as influence the structure and functioning of biological communities and the provision of ecosystem services [28,29].

Defining the targets of restoration, i.e., which measurable ecological goals are expected to be met once the restoration scheme is completed, is a critical step in the evaluation of the success of interventions [30]. Traditionally, targets are defined in terms of abiotic conditions of soil, water and sediments, or vegetation structure and composition that characterise habitats subjected to restoration [31]. However, in recent years, a wider approach emerged in restoration ecology, emphasising the importance of including functional attributes of re-created habitats (e.g., their trophic role for faunal communities and their ability to support overall biodiversity) among the measurable targets of restoration [32–34]. Furthermore, from a methodological perspective, restoration ecology is rapidly starting to incorporate predictive approaches for the evaluation of project performances [35]. Employing forecasting techniques would indeed contribute to provide robust and less uncertain assessments of restoration success, for instance, by predicting in advance the structure of biological communities prior to the end of habitat re-creation actions, [36] hence allowing for an adaptive managemen<sup>t</sup> of conservation e fforts [37,38].

Defining realistic targets and evaluating restoration success can be particularly di fficult in transitional water ecosystems. Here, the high levels of natural stress and the hysteresis exhibited by many ecological components after restoration often mask the response of the ecosystem to managemen<sup>t</sup> and conservation measures [16,24,39,40]. Due to their central role in ecosystem functioning, nekton fauna in transitional waters is widely employed as indicator of ecological health, being also included among the biological elements to be evaluated for the assessment of the ecological status of European transitional water bodies under the Water Framework Directive [41–44]. Estuarine and lagoon nekton communities are also considered in studies investigating the e ffects of restoration measures, including those involving the managemen<sup>t</sup> of salinity regimes in hydrologically impaired water bodies [19,20,29,45–48].

In the Mediterranean basin, the intense human development occurred in the last centuries and the subsequent alterations of hydrodynamics and sediment balances determined major losses of transitional water habitats in estuaries and coastal lagoons [49–54]. The Venice lagoon (northern Adriatic Sea, Italy) in particular, experienced centuries of land claim and decreased freshwater and sediment inputs from the mainland due to historical diversion of major rivers [55,56]. During the last century, the higher hydrodynamic energy caused by jetty construction at the sea inlets and channel dredging further enhanced the morphological and hydrological alterations [57,58]. Overall, this determined the loss of extensive portions of saltmarshes and intertidal flats, and the almost complete disappearing of reedbed and oligo and mesohaline conditions that originally dominated the interface between the lagoon and the drainage basin [49,50,59,60].

In the Venice lagoon, a restoration project (LIFE Lagoon ReFresh; www.lifelagoonrefresh.eu) is planned to establish a salinity gradient in a currently euhaline shallow water area, by connecting the basin to an adjacent river course and creating a freshwater inflow up to 1000 L s<sup>−</sup>1. In addition, a series of morphological interventions are planned in the area. These include the arrangemen<sup>t</sup> of biodegradable modular elements and the transplantation of reed (*Phragmites australis* (Cav.) Trin. ex Steud.) and submerged angiosperms (*Ruppia cirrhosa* (Petagna) Grande and *Zostera noltei* Hornem.), which are expected to slow down the dispersion of freshwater and control turbidity and nutrient peaks during river overflows [61]. The creation of a new freshwater input, scheduled for autumn 2019, is expected to restore many of the transitional attributes in the lagoon area. In particular, the project foresees the enhancement of habitat functionality for nekton fauna [61].

This paper aims at defining the structure of nekton assemblage to be expected after the restoration of the salinity gradient in an inner, euhaline area of the Venice lagoon, following the predictive approach proposed by Scapin et al. [36] for seagrass nekton communities. The work is structured into two phases: (i) an assemblage-level model was calibrated, in order to explain nekton variability with temporal, environmental and habitat factors characterising sites located along natural salinity gradients in the Venice lagoon; (ii) a set of target scenarios were defined in terms of salinity and other environmental factors that are expected at the end of the restoration process. The taxonomical and functional structure of nekton assemblage expected according to such scenarios was subsequently predicted, using the model developed in the first phase.

### **2. Materials and Methods**

The Venice lagoon is the largest Mediterranean coastal lagoon, with a surface of approximately 550 km2, and is characterised by a microtidal regime, experiencing a tidal range of ±0.50 m during spring tides [62]. Two main watersheds are present in the lagoon, identifying three large sub-basins (northern, central and southern) [63]. Each sub-basin exchanges water with the sea through an inlet. The lagoon is mostly composed of shallow water areas, with an average depth of 1.2 m [64], which are intersected by a network of deeper channels leading inwards from the inlets and branching inside each sub-basin [63,65]. The lagoon is also characterised by several freshwater inputs from the drainage basin, the most consistent of which are located in the northern sub-basin [66] (Figure 1).

Shallow water areas are therefore characterised by strong gradients in environmental conditions such as salinity, dissolved oxygen, turbidity, trophic status and sediment granulometry, these being driven not only by natural processes, but also by multiple anthropogenic pressures [49]. Overall, salinity levels range from polyhaline (18–30) to euhaline (>30) conditions, although values can reach 5 in areas closer to freshwater inputs, depending on the intensity of river discharges [66]. This heterogeneity contributes to creating a complex mosaic of islands, saltmarshes, mud and sand-flats, seagrass meadows and man-made structures. Saltmarshes and intertidal and subtidal flats dominate the mosaic of shallow waters in the inner lagoon areas. Seasonal beds of macroalgae often occur on flats, and sparse patches of *Zostera noltei* and *Ruppia cirrhosa* can be present along marsh edges and in marsh creeks. Reedbed is now rare, and limited to areas more directly influenced by freshwater [61]. The lagoon area that will be subjected to the restoration of freshwater inflow is a mostly shallow, inner portion of the northern sub-basin. It is characterised by high water residence times (>20 days; [63]) and euhaline conditions, with salinity levels often higher than 30 PSU, and only occasionally lowered by intense rainfall events [61,66].

In this study, nekton and environmental sampling was carried out in 39 shallow water sites in the three sub-basins. The sites were located along the major salinity gradients, either in confined saltmarsh creeks or at marsh edges exposed to shallow flats, and encompassed the range of environmental and habitat variability that characterises the inner areas of the lagoon (Figure 1).

Nekton and environmental data were gathered from 10 years of surveys in the Venice lagoon (between 2004 and 2018; [44,67–70]) and included observations performed either in spring, summer or autumn depending on the specific survey. A total of 179 observations were included in the dataset. Nekton sampling was carried out by seine netting, following the protocol described in Franco et al. [67]. The specimens were identified at the species level, and total biomass (g) per species was registered. The data were standardised per area unit (g 100 m<sup>−</sup>2), allowing comparison between samples. Together with nekton sampling, water temperature (◦C), salinity (PSU), dissolved oxygen (DO, percentage of saturation) and turbidity (FNU) were measured with a multi-parameter probe, and the presence of macroalgae was recorded. A value of sediment grain size (percentage of sand in the 10 cm surface layer) was finally associated to each sampling site using data from previous studies [71–73].

**Figure 1.** The Venice lagoon and location of sampling sites. The sites are located either in saltmarsh creeks (red dots) or at saltmarsh edges (green). The dashed lines indicate the approximate positions of the main watersheds. The distribution of the saltmarsh (dark grey), lagoon channels (light blue) and main freshwater courses flowing into the lagoon (blue) are also shown. The red arrow indicates the location of the planned river diversion into the lagoon planned under the LIFE Lagoon ReFresh project.

Field data were employed in a model framework to identify factors driving nekton variability and then predict assemblage characteristics under scenarios of salinity reduction, following the approach proposed for the restoration of seagrass meadows in the Venice lagoon by Scapin et al. [36]. Negative binomial generalised linear models (GLMs) were fitted to biomass of each species contributing to 98% of total assemblage biomass. A set of different model formulations was considered, taking into account different combinations of temporal, geographical, and environmental predictors. Five model categories were included, investigating the following hypotheses (Table 1): None of the predictors considered affect the response variable (null model: category m0); the response variable is affected by the temporal factor only (seasonal and inter-annual variability; category m1); the response variable responds to both temporal factors and the sub-basin (category m2); the response variable is affected by temporal factors, the sub-basin and environmental characteristics of water and bottom surface (category m3); the response variable responds to temporal factors, the sub-basin, environmental characteristics and location (the latter specified either as saltmarsh creek or edge) (category m4). The relative influence of each predictor on variability of nekton assemblage was hence evaluated by comparing the different model formulations (Table 1) in a hierarchical framework [36,41]. Following the approach of the

*manyglm* function contained in the *mvabund* software package [74], the inference was carried out at the assemblage level by combining species-specific results in a global analysis. The significance of the contribution of additive variables to the simpler model was assessed by means of likelihood ratio tests with 1000 bootstrap iterations. This allowed to investigate a series of hypotheses on the different contribution of each predictor on the overall assemblage variability. Test t1 tested the hypothesis that the temporal factor would improve the null model; test t2 tested the hypothesis that sub-basin would improve the temporal model; tests belonging to category t3 tested the hypothesis that each water parameter and bottom characteristic would progressively improve the previous model; test t4 tested the hypothesis that location would improve a model already taking into account all the previous predictors (Table 1). The model formulation resulting from the outcomes of the tests (i.e., that including only relevant predictors) was then selected, and employed in the second phase of the work to predict the expected nekton assemblages.


**Table 1.** Generalised linear models (GLMs) formulations considered in this study, and a summary of likelihood ratio tests performed between pairs of models. For each model comparison, the predictors being tested are specified.

The predictive capability of the selected model was assessed by means of Spearman's r coefficients, calculated between observed and predicted biomass values by means of a k-fold cross-validation (k = 5, [75]). The coefficients were computed for the whole assemblage (i.e., using species, sampling sites and dates as replicates) and for each species separately.

Predictor coefficients estimated by the selected model were used to quantify the magnitude and the sign (either positive or negative) of species response to physico-chemical variables and location (coefficients for seasons, years and sub-basins not shown). In order to provide an assessment of the response at the assemblage level, the absolute (i.e., without sign) values of estimated coefficients for each predictor were averaged among species, using the mean species biomass as weight.

Following the approach of Scapin et al. [36], the values of predictor variables were defined in a set of scenarios, as expected at the end of the restoration scheme. This allowed the prediction of the assemblages expected after establishing the freshwater input and creating the salinity gradient in the project area.

Three restoration scenarios were included, accounting for three different target levels of salinity reduction: 10, 18 and 25 PSU, as expected at the end of the project (see also Figure A1 in Appendix A, [61]). A fourth scenario was taken into account, describing the current salinity conditions in the project area (ca. 30 PSU; [61]). All the scenarios were based on the northern lagoon sub-basin (i.e., where restoration will take place) and in all the scenarios, levels of environmental parameters were set as the inter-annual average values measured during each season in northern sub-basin sites employed in the calibration phase. The year was set as the average sampling year considered in calibration. Since saltmarsh creeks are not expected to be a prominent habitat feature in the restored area in the near future [61], in all the scenario explorations, predictions were carried out for the exposed marsh edge location. The scenarios were defined for spring, summer and autumn separately (Table 2).

Species predicted with low accuracy (i.e., species associated with a Spearman's cross-validation coefficient lower than 0.25) were excluded from the predicted assemblage. In order to provide a functional assessment of the expected assemblage, species were subsequently grouped into functional guilds. Both ecological guilds, summarising the different use of transitional water habitats by species, and feeding guilds, grouping species with similar food targets and foraging strategies, were considered. Predicted species densities were then aggregated accordingly. Eight ecological guilds (estuarine use functional guilds, EUFGs) were taken into account, adapting the classification approach of Potter et al. [76] to the Venice lagoon: Solely estuarine resident species (ESs), found exclusively within the lagoon ecosystem; estuarine resident species (ES), spending all or most of their life cycle within the lagoon, but represented also by marine populations; marine estuarine-dependents (ME-D), marine-spawning species that require transitional water habitats during the juvenile stages; marine estuarine-opportunists (ME-O), marine-spawning species that regularly enter the lagoon but can alternatively use other coastal habitats; marine stragglers (MS), stenohaline marine species irregularly found within the lagoon in areas most influenced by the sea; freshwater stragglers (FS), stenohaline freshwater species that occur occasionally in the lagoon near river mouths; catadromous (C), species entering the lagoon during periodic migrations from marine spawning areas to freshwaters; anadromous (A), species entering the lagoon during periodic migrations from freshwater spawning areas to the sea. Seven feeding guilds (feeding mode functional guilds, FMFGs) were taken into account, adapting the classification of Franco et al. [77] to the Venice lagoon: Detritivores (D), feeding on small organisms and organic matter associated to the substratum; microbenthivores (Bmi), feeding on benthic fauna smaller than 1 cm; macrobenthivores (Bma), feeding on benthic fauna larger than 1 cm; hyperbenthivores-zooplanctivores (HZ), feeding either on small (<1 cm) hyperbenthos or zooplankton; hyperbenthivores-piscivores (HP), feeding either on large (>1 cm) hyperbenthos or fish; omnivores (OV), which ingest both plant and animal material; planktivores (PL), feeding predominantly on zooplankton and occasionally on phytoplankton. Since a species could be allocated to multiple feeding guilds, the contribution of each species to a guild was expressed as a proportion (0 to 1), by identifying the importance of different food resources within the diet based on literature [78] and available data for the Venice lagoon [41,79].


