Large-scale restoration of Mediterranean macroalgal forests is set to accelerate under new policy commitments, yet many initiatives rely on high initial settlement densities despite limited understanding of how density regulates early recovery. If early demographic bottlenecks constrain survival, growth or reproduction, restoration strategies based on maximising numbers may undermine success. We experimentally manipulated initial settlement density of the canopy-forming brown alga Cystoseira compressa across four densities (10, 100, 250, and 500 individuals) spanning natural to highly elevated levels. Recruits were monitored every 3 months for 1 year in the field to quantify survival, growth, reproductive development and the timing and intensity of density-dependent bottlenecks. Survival declined across all densities but showed strong density dependence. The lowest density maintained high survival throughout the experiment, whereas intermediate and high densities experienced rapid early mortality. At the highest densities, survival collapsed within the first 2–3 months, indicating acute early bottlenecks rather than gradual mortality. Mean final densities at the end of the experiment ranged between 0.19 ind. cm ^-2 and 0.44 ind. cm ^-2. Density strongly constrained individual growth. The lowest density rapidly progressed into larger size classes and were the only density where individuals reached reproductive maturity. Higher densities suppressed growth of surviving individuals, resulting in persistent confinement to small size classes and preventing reproductive development despite continued survival. Quantification of survival derivatives revealed that both the timing and intensity of bottlenecks scaled predictably with initial density. Bottlenecks were absent at lower densities, prolonged at intermediate densities and short but severe at the highest densities, demonstrating that early recovery trajectories were density-driven rather than time-dependent. This single-species and single-site experiment provides evidence that high-density restoration strategies accelerate mortality, suppress growth and delay reproduction, whereas low-density approaches promote faster demographic progression and resilience. Under the conditions tested here, focusing on growth and reproductive output, rather than maximising initial numbers, reduces the need for large propagule supply and extensive receptacle production, directly supporting the conservation of donor populations while maintaining restoration effectiveness.

Collection site and culture

Saint-Jean-Cap-Ferrat is a rocky peninsula on the south-eastern French Mediterranean coast, located approximately 6 km east of Nice and 12 km west of Monaco (Figure 1 in associated manuscript). The collection site, Pointe du Colombier, is characterised by shallow subtidal rocky habitats (0.1–1 m depth) typical of Mediterranean Cystoseira forests, dominated by Cystoseira compressa, Ericaria crinita, C. compressa f. rosetta, and E. amentacea.

Approximately 200 g of fertile branches of C.compressa were collected from the donor population at Pointe du Colombier on 27 May 2024 (Figure 1a). As an ecosystem engineer, C. compressa is an ideal model species for restoration experiments, being relatively tolerant, easy to work with it and has been shown to increase the chance of restoration success (Susini et al., 2007). To avoid inadvertently detaching individuals, the forest was approached by swimming (not walking) and branches were carefully removed using scissors. Branches were then transported to the laboratory at Université Côte d’Azur (Nice) in sealed plastic bags without seawater, placed in a cooler to maintain dark and cold conditions during transport. Upon arrival, fertile material was stored at ~4 °C for 15 h to promote zygote liberation following established protocols (see Cebrian et al., 2021). To assess natural settlement densities of C. compressa at Pointe du Colombier, small experimental units (n = 10; area = 7.07 cm²; hereafter EU) were deployed beneath existing fertile canopies. Substrate was cleared using a wire brush and units were affixed to the rock using marine epoxy (IVEGOR Multi Putty) and left for 1 month.

Experimental aquaria were set up in a closed-circuit system supplied with filtered natural seawater (200 µm mesh) and continuously aerated using air pumps. Lighting conditions reflected the natural seasonal photoperiod, with large EUs (area = 36.5 cm²) exposed to ambient natural daylight. Temperature conditions were maintained between 18–20 °C. Fertile branches were evenly distributed across 16 tanks (internal dimensions: 18 × 24 × 12 cm; volume: 3.5 L), using a submerged mesh hammock positioned above two EUs per tank (12.5 g per tank; Figure 1b). Gamete release commenced on 28 May 2024 and continued until 31 May 2024. Following release, mesh hammocks were removed, branches discarded and EUs were placed into new tanks of filtered natural seawater. EUs used in the ex-situ field experiment (n = 20) were assigned to one of four predefined settlement densities; low density: 10 ind EU ^-1 (hereafter LD), semi-intermediate density: 100 ind EU ^-1 (hereafter SID), intermediate density 250 ind EU ^-1 (hereafter ID) or high density: 500 ind EU ^-1 (hereafter HD)) on 3 June 2024. Densities were manipulated under a stereomicroscope using fine forceps to selectively remove excess individuals until the target densities were achieved, while maintaining an even spatial distribution of remaining individuals across each unit to minimise clumping. Each unit was tagged and given a unique ID number. Following settlement, experimental units were maintained under stable conditions for three weeks, with seawater replaced every 3–4 days.

Study site and monitoring

The Pointe du Colombier site and surrounding rockpools experience regular and intense grazing pressure, primarily from Sarpa salpa (Gianni et al., 2018). In addition, Saint-Jean-Cap-Ferrat is considered a tourist hotspot and shallow rocky habitats are subject to frequent physical disturbance through trampling. To minimise these external pressures an alternative monitoring site was selected.

Cap d’Antibes is a rocky peninsula on the south-eastern French Mediterranean coast, located approximately 20 km southwest of Nice and 10 km east of Cannes (Figure 1c). The selected rockpool is semi-enclosed, while still permitting regular water exchange but protected from the threat grazing as to isolate and test the impact of density. The rockpool measures 7.71 m × 6.04 m and is relatively flat, with a water depth ranging between 0.2 and 0.5 m. The benthic community within the pool is dominated by E. crinita and C. foeniculacea, with C. compressa and E. amentacea present in the surrounding area.

EUs (n = 20) were deployed in the field using the same attachment method described above and were placed randomly within the rockpool on 24 June 2024. Monitoring was conducted over a 13-month period, from June 2024 to July 2025. To facilitate repeated monitoring, the rockpool and positions of each unit were mapped prior to the start of observations. Monitoring was conducted at three-month intervals to quantify the density, survival and growth of C. compressa. Survival was assessed as the number of individuals present on each EU, while growth was quantified as the length (cm) of the longest axis of five randomly selected individuals per EU. Finally, reproductive activity was assessed by recording the presence and extent of receptacle development following the categorical protocol of Rindi et al. (2023). Individuals were classified as non-fertile (NF) or assigned to one of four fertility classes based on the proportion of branches bearing receptacles: < 25 % (F1), 25–50 % (F2), 50–75 % (F3), and > 75 % (F4).

Data analysis

Survival was analysed using a binomial generalized linear mixed-effects model (GLMM) to quantify the influence of initial establishment density and time on post-settlement survival. Survival at each sampling time was expressed as the number of the individuals alive relative to the number initially established per EU. Initial densities (10, 100, 250, and 500 ind EU ^-1), time (sampling month) and their interaction were included as fixed effects to test whether survival trajectories differed among densities over time. EU ID was included as a random intercept to account for repeated measurements. Time was treated as a categorical variable to avoid imposing assumptions of linear decline. The initial sampling month (0) was excluded from the analysis as survival was fixed at 100 % by experimental design. Models were assessed using Type III Wald chi-square tests and where significant effects were detected, survival probabilities and 95 % confidence intervals were estimated. Pairwise comparisons among densities were conducted within each time point using odds ratio, with p values adjusted for multiple comparisons using the Holm method.

Changes in size among densities was assessed using non-parametric statistics. At each sampling time (3, 6, 9, 12 months), individual sizes (cm) were compared among densities (10, 100, 250, and 500 ind EU ^-1) using Kruskal-Wallis tests as size data violated assumptions of normality and homoscedasticity. When tests were significant, pairwise post-hoc comparisons between densities were conducted using Dunn’s tests with Holm correction for multiple comparisons. Growth rates (cm month ^-1) were calculated for each density between successive sampling times and analysed using a linear mixed-effects model with density, interval midpoint and their interaction as fixed effects and EU identity as a random intercept. Fixed effects were evaluated using Type III Wald chi-square tests and differences in density were assessed using Holm-adjusted pairwise comparisons of model-estimated marginal means.

To quantify bottleneck timing and intensity, we analysed changes in survival over time using the numerical derivative of smoothed survival trajectories. Mean survival proportions were calculated for each density (10, 100, 250, and 500 ind EU ^-1) at each sampling time (3, 6, 9, 12 months). Generalised additive models (GAMs) were then fitted separately for each density, with time included as a smooth term, allowing for non-linear survival dynamics without imposing parametric assumptions on the shape of decline. The first derivative of survival with respect to time (dS/st) was estimated numerically for each density. This derivative represents the instantaneous rate of change in survival per month, with values close to zero indicating little mortality and increasingly negative values indicating faster rates of mortality. Bottlenecks were identified using a rule-based, two-step criteria. First, a minimum mortality threshold of −0.05 per month (equivalent to ≥ 5 % survival loss per month) was imposed to ensure that only ecologically meaningful mortality were considered. Bottlenecks were only classified when the most negative rate of survival exceeded this threshold. Second, for densities meeting this criterium, bottleneck window was defined as the period during which dS/dt remained below 50 % of the most negative slope. Bottleneck start and end times were identified as the upper and lower bounds of this period. This approach allowed for bottlenecks to be identified objectively and consistently across densities ensuring they reflected sustained periods of intense, density-dependent mortality rather than brief stochastic fluctuations or background decline.

All statistical analyses were conducted in R (R Core Team, 2021). Survival proportions and growth rates were analysed using generalised and linear mixed-effects models fitted with package lme4. Estimated marginal means were obtained using the package emmeans. Differences in size distributions were evaluated using Kruskal–Wallis tests followed by Dunn post hoc tests with Holm correction. Density-dependent survival bottlenecks were quantified by fitting generalised additive models (GAMs) to smoothed survival trajectories (mgcv).