Capraia, the third largest island in the Tuscan Archipelago with an area of 19.2 km², lies closer to the Corsican coast than to the Tyrrhenian mainland (Fig. 1). The island is mostly mountainous and has a Mediterranean climate, characterized by hot, dry summers and mild, rainy winters (Foggi and Grigioni 1999). Its vegetation is largely dominated by evergreen shrubland, where Erica arborea and Arbutus unedo prevail, gradually giving way to more pioneer plant communities in the steeper and rockier areas (Foggi and Grigioni 1999). Forests of Quercus ilex are almost absent, except for a few small patches, likely due to centuries of continuous human disturbance (Foggi and Grigioni 1999). Despite this, Capraia retains high ecological value, hosting numerous endemic plant species as well as terrestrial habitats of notable conservation importance. Among these, two habitat types (3170* “Mediterranean temporary ponds” and 6220* “Pseudo-steppe with grasses and annuals of the Thero-Brachypodietea”) are listed as “priority” under the European Habitats Directive (92/43/EEC). The island is also included in both the Tuscan Archipelago National Park and the Natura 2000 network, designated as a Special Area of Conservation (SAC) and a Special Protection Area (SPA) (IT5160006). However, Capraia faces a serious threat from Opuntia stricta, introduced in the 1980s (Viegi and Cela Renzoni 1981; Guiggi 2008), which has since spread extensively. Today, it covers around 70 hectares (3.6% of the island), endangering almost half of the habitats of conservation concern as defined by the Habitats Directive. In the absence of natural enemies or control measures, the species continues to expand across natural and semi-natural areas, exacerbating its impact on ecologically valuable habitats.

In particular, this study focuses on the following target habitats: 1240 “Vegetated sea cliffs of the Mediterranean coasts with endemic Limonium spp.”, 3170* “Mediterranean temporary ponds”, 5230 “Low formations of Euphorbia close to cliffs”, 5330 “Thermo-Mediterranean and pre-desert scrub”, 6220* “Pseudo-steppe with grasses and annuals of the Thero-Brachypodietea”, and 8220 “Siliceous rocky slopes with chasmophytic vegetation”.

Sampling was conducted in May 2024 and May 2025 in the north-eastern part of the island, where Opuntia stricta is significantly dense (Misuri et al. 2024) and where removal interventions will take place. As reported by Misuri et al. (2024), in fact, Opuntia stricta, with over 70 ha, is the most widespread alien species that extends significantly outside the town in areas with natural vegetation, reaching the National Park territories in small and scattered patches. The estimated density of these patches is derived from an empirical estimate obtained from field observations by Misuri et al. (2024), based on the number and width of Opuntia clumps per area.

We employed a hierarchical sampling design. The first level consisted of a random sampling, by which a total of 25 permanent square macroplots (100 m² each) were randomly selected within the study area in 2024 (Fig. 1). This was combined with a systematic within-plot sampling at the second level, where two subplots (4 m² each) were established, one in the north-eastern and the other in the south-western corner of each macroplot. For each macroplot, we recorded the cover of the target habitats — 1240 Vegetated sea cliffs of the Mediterranean coasts with endemic Limonium spp., 3170* Mediterranean temporary ponds, 5320 Low formations of Euphorbia close to cliffs, 5330 Thermo-Mediterranean and pre-desert scrub, 6220* Pseudo-steppe with grasses and annuals of the Thero-Brachypodietea, and 8220 Siliceous rocky slopes with chasmophytic vegetation — as well as the cover of Opuntia stricta. For each subplot, we recorded the occurrence and cover of plant species as a direct percentage estimation (with a minimum value of 0.1%), and collected any plant species that could not be identified on the field. After conducting fieldwork, we identified the samples collected in the laboratory using a stereoscope, in accordance with Pignatti et al. (2017–2019). We also updated the nomenclature according to the Portal to the Flora of Italy (2025).

To identify the plant species associated with the reference physiognomic combination for each target habitat, we consulted the Italian Interpretation Manual of the 92/43/EEC Habitats Directive (Biondi et al. 2010). Furthermore, we checked the typical species within the “Handbook for monitoring species and habitats of community interest (Council Directive 92/43/EEC) in Italy: habitat types” (Angelini 2016). We therefore created a table in which we counted the species from the physiognomic combination and typical species in the surveys for each target habitat.

Moreover, to provide a broad classification, each macroplot was assigned in the field to one of the following three vegetation macrocategories based on visual inspection:

To explore the patterns of cover of Opuntia stricta and the variations in habitat cover across different vegetation macrocategories and over time, we adopted a repeated measurements framework using Linear Mixed-Effects Models (LMMs) with macroplot identity treated as a random effect to account for temporal correlations in repeated measurements. Opuntia stricta cover was analyzed in function of sampling year, vegetation macrocategories, and their interaction. A second LMM was used to study the cover of Natura 2000 habitats in function of vegetation macrocategories, habitat type, sampling year along with their interactions. The statistical significance of the fixed effects was assessed using a Type III Analysis of Variance (ANOVA) with Satterthwaite’s approximation for denominator degrees of freedom. Moreover, we used an alluvial diagram to represent the overall compositional structure of the vegetation community, summarizing the flows between macrocategories and Natura 2000 habitats. In this case, cover values were averaged across years (also considering the non-significance of the factor Year).

To explore the vegetational context and to visualize patterns in plant community composition of the sampled plots, we performed a Non-metric Multidimensional Scaling (NMDS) ordination based on the Bray-Curtis dissimilarity index, using species cover data from all subplots. Three dimensions (k = 3) were used, as the two-dimensional solution (k = 2) resulted in a high stress value (stress = 0.22). Differences in community composition among a priori-defined vegetation macrocategories, as well as the effects of sampling year were tested using permutational multivariate analysis of variance (PERMANOVA). To further interpret the ordination axes and vegetation macrocategories, sampling year and cover in Natura 2000 habitat recorded in the field, we fitted these variables onto the NMDS space using the envfit function (vegan package). Continuous variables were fitted as vectors, while categorical factors were represented by their centroids. Only relationships significant at p < 0.05 were displayed in the final ordination plot.

Finally, we performed a golden test to assess the reliability of the field-based habitat classification against the floristic data, thereby assessing the diagnostic value of the reference species combinations for Natura 2000 habitats. Habitat cover data for each macroplot were converted into a binary presence/absence matrix, defining two groups of plots per habitat type (“present” vs. “absent”). For each plot, two metrics were calculated based on species belonging to the reference combinations: (1) the total number of diagnostic species and (2) their cumulative cover. Differences in these metrics between the two groups were tested using non-parametric Wilcoxon rank-sum tests.

The statistical analyses were conducted in the R environment (R Core Team 2025). The vegan package (Oksanen et al. 2022) was used for all multivariate community analyses, including NMDS ordinations, PERMANOVA, and vector fitting. The package lme4 (Bates et al. 2015) was used to fit Linear Mixed-Effects Models, while the package lmerTest (Kuznetsova et al. 2017) provided p-values for fixed effects. The package tidyr (Wickham and Henry 2020) was used to reshape the dataset, and the package ggplot2 (Wickham 2016) to create all plots. The package ggalluvial (Brunson and Read 2023) was used to generate alluvial diagrams representing vegetation composition, and cowplot (Wilke 2019) combined multiple plots into a single figure.

In our monitoring, we recorded Opuntia stricta occurrence in almost all macroplots and subplots, in some cases with relevant cover (higher than 50%). However, the observed patterns in cover across vegetation macrocategories suggest differential habitat susceptibility to invasion and a certain degree of spatial heterogeneity in its distribution. In fact, the higher mean cover recorded in Coastal vegetation and Low maquis, combined with the slight increase over time in coastal areas, indicates that these habitats may provide particularly favorable conditions for the establishment and expansion of the species, in accordance with the findings of Misuri et al. (2024). In fact, open environments and disturbance may facilitate cacti colonization and growth (Shackleton et al. 2017; Sheehan and Potter 2017). These findings are also supported by the NMDS analysis, which suggested an association between this invasive species and open coastal habitats. Furthermore, in accordance with Padrón et al. (2011) and Gómez-Bellver et al. (2020), the occurrence of opuntioid cacti was reported in rocky coastal habitats and in Mediterranean scrub mixed with abandoned fields.

However, the consistently low cover observed in Low garrigue suggests that this vegetation macrocategory might be less suitable for the invasion. This evidence seems in contrast with the above findings for open environments, in fact open spaces are certainly a defining feature of low garrigue. This result is probably linked to several factors, including less favorable ecological characteristics, greater competitive pressure from native species, the random sampling design used, and how the species disperses and spreads on the island. Further investigation is required to develop a more comprehensive understanding of this dynamic.

By focusing on the relationship between vegetation cover and the Natura 2000 habitat, our results revealed a significant heterogeneity and clear distributional patterns. Coastal vegetation presented a highest mean cover associated with habitat 5320 and 1240, indicating a close correspondence. Likewise, the specific composition demonstrated a notable similarity between Coastal vegetation and habitats 5320 and 1240, indicating a strong affinity between them. Conversely, Low garrigue and Low maquis generally displayed lower overall mean habitat cover but are widespread across several habitats. In particular, Low garrigue is mainly represented by the habitat 6220* and 8220 cover, while Low maquis exhibited a higher mean cover for the habitat 5330. Similarly, the species composition of both these two vegetation macrocategories showed a partial overlap reflecting more gradual ecological transitions between these habitats.

The connections established between vegetation macrocategories and Natura 2000 habitats enable us to identify the dominant structural components. Understanding these associations is crucial for conservation planning and management, as it enables us to identify broader vegetation categories that encompass multiple habitats. This allows us to determine where to concentrate our conservation efforts.

From the floristic-vegetation surveys conducted on the 50 subplots, we recorded a total of 178 species over the two years of monitoring, approximately 25% of the total flora checked by Foggi et al. (2001). This result can be considered remarkable given the limited number of surveys and the few macrocategories of vegetation explored. In addition, the complex vegetational and spatial dynamics of Capraia’s plant landscape must also be considered. The island’s plant communities are formed from mosaics of whose vegetation types are often mixed with one another, making Capraia highly valuable in terms of conservation (Foggi and Grigioni 1999).

Among the endemic species reported by Foggi et al. (2001), during the two years of monitoring, we found Galium caprarium, Romulea insularis, and Limonium caprariae. Furthermore, Limonium caprariae belongs to the reference physiognomic combination of habitat 1240 “Vegetated sea cliffs of the Mediterranean coasts with endemic Limonium spp.”, as a species of the genus Limonium, and is considered a typical species of this habitat (Biondi et al. 2010; Angelini 2016).

By focusing on the species of the reference physiognomic combinations, the high number of species found in habitat 6220* compared to other habitats is not surprising, given that this habitat is inherently particularly rich in species (Biondi et al. 2010). These communities are mainly established in the empty spaces between shrubs, as well as in the most degraded microelophytic ephemeral grasslands. They are usually small in size and characterized by annual graminoid therophytes, most of which belong to the alliance Tuberarion guttatae (Foggi and Grigioni 1999). Often, the xerotolerant annual species of these communities invade temporary ponds as they dry up, creating contact between different plant communities and habitats.

For the habitat 3170*, we did not find a large number of species from the reference physiognomic combination, probably due to the ephemeral nature of the plant community and the limitation of the surveys. However, we recorded other entities characteristic of late winter microelophytic ephemeral meadows (class Isoëto-Nanojuncetea), such as Romulea insularis, Romulea columnae, Bellis annua, and Lotus angustissimus, in accordance with Foggi and Grigioni (1999).

Regarding the coastal vegetation represented by habitats 1240 and 5320, we found communities consisting of Helichrysum italicum and Jacobaea maritima in spatial contact with scrub formations on one side and aeroaline chasmophytic communities of association Crithmo-Limonietum contortiramei on the other. These communities are underestimated compared to those of other macrocategories of vegetation, partly due to the sampling design and partly due to the difficulty, if not impossibility, of accessing these areas.

On the other hand, habitat 5330 is fairly well represented in our monitoring at Capraia. It is characterized by patches of Euphorbia dendroides, which mainly grow in steep rocky areas, sometimes merging with the Helichrysum belt near the sea, especially on southern exposures and eastern slopes. We often recorded its occurrence in with Cistus monspeliensis, Teucrium marum, Pistacia lentiscus, Myrtus communis, Salvia rosmarinus, Erica arborea, and in accordance with Foggi and Grigioni (1999), occasionally with Olea europaea.

For habitat 8220, we did not record any species from the reference physiognomic combination. However, we did record other entities (Umbilicus rupestris, Polypodium cambricum, and Selaginella denticulata), which, according to Foggi and Grigioni (1999), are included in the xerophytic, chasmophytic vegetation characteristic of the alliance Asplenio billotii-Umbilicion rupestris communities and related impoverished phytocenoses. In addition, we found Galium caprarium, characteristic of the association Linario caprariae-Umbilicetum rupestris of chasmophytic vegetation (Foggi and Grigioni 1999).

In any case, we found a clear and consistent association between the species richness and total cover of the reference physiognomic combination across Natura 2000 habitats (except for habitat 8220). Only in a few cases this relationship was not evident. For example, in habitat 3170*, species cover of the reference physiognomic combination was similar among plots with and without the habitat, suggesting the possible influence of other factors, such as the dominance of ephemeral species and seasonal changes. In other instances, we observed high values of species richness and cover in plots where the corresponding habitat was not formally identified. This discrepancy may stem from both analytical and interpretative factors. Indeed, the concept and delineation of habitats are based not only on the occurrence of species belonging to the reference physiognomic combination, but also on the complex interactions between abiotic and biotic factors that shape community structure and function (The Council of the European Communities 1992; Angelini 2016; IPBES 2019).

Certainly, in some cases, the absence of species characteristic of the reference physiognomic combination and plant community of the habitat from our records does not reflect their actual absence in the study area, but rather their omission due to not being included within the 4 m² subplot. While our sampling design enabled the detection of a considerable number of species, it also presented certain limitations, particularly for some habitat types. In fact, this is not a stratified sampling design where we have a homogeneous distribution of macroplots among vegetation macrocategories and habitats, but rather a random sampling design at the first level (macroplots), combined with a systematic within-plot sampling at the second level (subplots). Nevertheless, our objective was not to produce a detailed phytosociological description of the study areas, but to assess and delineate the conservation status of the habitats threatened by the invasive species, Opuntia stricta. We note here that in our assessment, we also attempted to use the framework of “typical species” as defined by the Habitats Directive (92/43/EEC) and listed in the Italian Manual for Habitat Monitoring (Angelini et al. 2016). However, we encountered several issues. Although typical species are theoretically intended to indicate the conservation status of habitats, we found that they were reported only for a limited number of habitat types in Angelini et al. (2016), and that, in many cases, the listed taxa did not appear particularly responsive to environmental changes. Moreover, for several habitats (e.g., 6220*, 3170*, and 8220), the manual omits typical species, recommending their identification at a regional scale due to high floristic variability. These limitations reflect the broader conceptual and practical difficulties in defining and applying the concept of typical species, as repeatedly discussed in the scientific literature (Bonari et al. 2021). Nevertheless, in the context of post-restoration monitoring, identifying reliable typical species could be extremely valuable. This is particularly relevant for LIFE projects and other conservation initiatives, where one of the main indicators of success is the measurable improvement of habitat conservation status — an evaluation that currently remains difficult, if not impossible, when relying on the existing framework of typical species.