The overall floristic pattern of the Algerian islands and islets was analysed through multivariate analyses, i.e. Principal Component Analysis (PCA) coupled with Hierarchical Clustering on Principal Components (HCPC), for two sets of variables. The first PCA compared sites described by their floristic composition (occurrence of plant species and subspecies in each island or islet). Presence-absence data were Hellinger-transformed to account for the double-zero problem (Borcard et al. 2018). The second PCA was applied to flora descriptors (chorological type, life form, seed dispersal syndrome, pollination syndrome, species richness), scaled to zero mean and unit variance. PCA and HCPC were performed using functions of the R package FactoMineR v.2.11 (Le et al. 2008). The HCPC function applied the Ward clustering method to the Euclidean distances between site scores across the first ten PCA axes.

A Redundancy Analysis (RDA) was used to explain flora descriptors by physiographic (elevation, area, distance from the mainland, SI, PAR, latitude and longitude) and biotic (human presence, number and density of yellow-legged gulls Larus michahellis) variables. RDA was performed using the rda function of the vegan package v.2.6-4 (Oksanen et al. 2022). A stepwise selection of explanatory variables was applied to retain the most parsimonious model, using the ordistep function of the same package. In addition, each explanatory variable was tested in separate RDAs using the vartest function (Borcard et al. 2018).

All analyses were performed in R v.4.3.2 (R Core Team 2023).

The HCPC analysis applied to the sites × species occurrence matrix allowed us to identify five main groups of islands and islets (Fig. 2, Table 3).

The two first components of the PCA represent 20.8% and the ten first components 61.4% of the total variance. Group 1 is differentiated by the presence of Mesembryanthemum crystallinum L., Fumaria munbyi, Lycium intricatum Boiss., and Malva durieui Spach (Fig. 3B) with often a high species richness (Fig. 4A). Groups 2 and 3 are characterized by the lowest species richness, with the presence of Mesembryanthemum nodiflorum L. and Suaeda vera Forssk. ex J.F.Gmel. (group 2), and of Malva arborea, and Senecio leucanthemifolius Poir. (group 3). Group 4 represents species-rich small islands with Limbarda crithmoides (L.) Dumort., Anthemis maritima, and Beta vulgaris subsp. maritima (L.) Arcang. (Fig. 3B). Group 5 features islands with the highest species richness, and the presence of Lotus ornithopodioides L., Solanum nigrum L., and Phillyrea latifolia L. (Supplementary material 1).

The five groups of islands and islets identified from the PCA scores (Fig. 3A) are characterized by contrasted environmental conditions (Fig. 4). Group 1 is composed of five large, including the three largest islands (Grande Habibas, Rachgoun, and Petite Habibas) (Fig. 4B), isolated (Fig. 4D) islands of the western Algerian costline, hosting a big gull population (Fig. 4H) with a frequent human presence. Group 2 is made of six relatively small islands and islets (Fig. 4B), located close to the mainland (Fig. 4D), with a low gull density (Fig. 4H). Group 3 contains the eight smallest islands with low human frequentation and the lowest concentration of seabirds (Fig. 4H, I). Group 4 comprises four islands moderately isolated from the coast, with low gull concentration. Group 5 corresponds to eight islands with the highest gull density (Fig. 4H).

No significant difference was shown between the five groups concerning steepness index and human presence (Fig. 4F, G, respectively).

The PCA of the matrix sites × flora descriptors, in combination with a HCPC (Fig. 5), allowed us to classify islands and islets into three main groups (Fig. 6A).

The two first components of the PCA represent 36.1% of the variance and the ten first 84.9%. The first axis is strongly positively correlated to the species richness (S), a zoochorous dispersal syndrome, the proportion of geophytes (Ge) and of introduced species, and negatively to the proportion of chamaephytes (Ch), Mediterranean (Medsl), and barochorous species (Fig. 6B). Axis 2 is negatively influenced by the percentage of entomogame species and of phanerophytes (Ph), and positively by the proportion of anemochorous species.

Group 1 is represented by eight islets or small islands (Fig. 7B), mainly located in front of the central and central-western coasts of the country (Aguéli, Hadjrat Ennos, Sekia, and Dziria), and a low gull concentration (Fig. 7H, I). It is characterised by a low species richness (mean = 8, Fig. 7A) and a higher proportion of chamaephytic, barochorous, and Mediterranean species.

Group 2 contains fourteen islands and islets. The main sites that best characterise this group are relatively small islands (Fig. 7B) with high proportions of entomogamous and phanerophytic species, a relatively low species richness (mean = 18, Fig. 7A), and a low gull density (Fig. 7H).

Group 3 includes eight large islands (Fig. 7B), which host the highest number of plant species (mean = 65, Fig. 7A), mainly composed of zoochorous and geophytic species, with the lowest proportion of Mediterranean species in the broad sense, the highest proportion of Eurasian-holarctic and introduced species, and a higher gull density (Fig. 7H, I).

The three groups differ markedly concerning the biotic variables (Fig. 7H, I), the islands in group 3 hosting the largest populations of gulls. No significant difference was shown between the three groups concerning elevation, isolation, perimeter/area ratio, steepness index, and human presence (Fig. 7C, D, E, F, G, respectively).

Taken alone, the density of gull colonies (dGull) and the island area were the variables that best explained the variation of the flora descriptor matrix (Table 3). After the stepwise selection of explanatory variables, a combination of four of them were retained in the parcimonious RDA model: gull density, area, latitude, and longitude. These variables explained together 26.6% of the variance of the flora descriptor matrix.

This RDA (Fig. 8) reveals that the segregation of the sites along the first canonical axis is influenced by the density of the gull population (dGull) and the area of the islands or islets. This main gradient isolates group 3 (large islands enriched with Eurasian and introduced species and with a high density of gulls) from groups 1 and 2 (smaller islands with a higher proportion of Mediterranean species).

On the other hand, the separation of sites along the second canonical axis is driven by their geographical location (longitude and latitude), irrespective to the main gradient separating the three groups. Interestingly, endemic, entomogamous, and autochorous or hydrochorous species occur mainly in the western and northern part of the study area (lower longitude East, higher latitude North), particularly in four islands (Sbiaat, Mokreum, Petite Habibas, and Île aux Rats).

The classification of the 30 Algerian islands and islets through the PCA highlighted the importance of physiographic variables: in western Algeria, islands are represented by the Habibas, Rachgoun, Mokreum, and Sbiaat islands, and are characterised by fairly large surfaces and marked isolation compared to the other islands and islets, a specificity associated with a biogeographical pattern characteristic of the Oranese sector, the Baetic-Rifan arc, and the Iberian-Mauritanian ensemble (Véla et al. 2013). In contrast, the majority of the islets or rocks located on the Central and Eastern coasts of Algeria are very close to the continent and they are characterized by a biogeography influenced by the Kabyle sector, the Algerian-Constantinian Tell, and the Tyrrhenian microplate complex. The same phenomenon occurs in the far east, according to a preliminary analysis of recent data by Hamel et al. (2023).

The morphology and the elevation of the studied islands are not very important factors for the islands’ classification; the majority of them are characterised by a more or less flat shape and low elevations. As a result, the physiognomy of the Algerian islands is quite peculiar compared to the small islands of Sardinia, for which Fois et al. (2016) demonstrated the importance of the elevation and the SI factor, which strongly contributed to the floristic structuration of these small islands. Indeed, Fois et al. (2016) refer to islands with much larger dimensions and elevations above 500 m a.s.l. The Algerian coast is characterised by a scarcity of islands and islets, compared to the other Mediterranean countries (Médail 2022), and in addition to their rarity, the few existing insular entities are in great majority very small islets of less than one hectare. In fact, when considering the physiographic variables and beside a longitudinal gradient, surface area and isolation (distance from the mainland) resulted to be the two most important determinants of the floristic composition of the Algerian islands.

As already pointed out by Benhamiche-Hanifi and Moulaï (2012), the flora of Mediterranean origin is dominant on the islands of Jijel and Bejaia, and this pattern was observed on all the studied islands, where the flora is largely dominated by the Mediterranean species (Supplementary material 2). The same characteristics were observed by Hamel et al. (2023) for the islands at the extreme east of the country. However, there is a west-east difference, with the Oranese islands being influenced by the western Mediterranean sub-domain, including the Ibero-Maghrebian complex, whereas the Kabyle and Numidian islands are influenced by the central Mediterranean sub-domain, including the Tyrrhenian complex (Véla and Pavon 2013; Hamel et al. 2023).

In the case of the Algerian and Tunisian islands, Véla and Pavon (2013) showed that a larger surface area does not necessarily mean higher richness, nor does reduced richness mean lower conservation interest, as in the case for Rachgoun Island (large but poor although of high conservation interest), and for the Islets of Garlic and Sahel (small but of high conservation interest). This dissymmetry between surface area, richness, and conservation interest is also confirmed at the extreme east of the country by Hamel et al. (2023). Due to their small surface area, the Algerian islands are not considered as regional hotspots; this is mainly due to the fact that the small surface area is insufficient to allow in situ speciation (Lomolino and Weiser 2001), and consequently these sites are very poor in exclusive endemics (Triantis et al. 2008). The Oranese islands belong to the regional hotspot of the Baetic-Rifan Arc and the islands of the coast east of Algiers belong to the regional hotspot Kabylie-Numidie-Kroumirie (Véla and Benhouhou 2007). Nevertheless, for the Habibas archipelago, plant endemics are shared with at least one site on the mainland (Africa and/or Europe) or with other archipelagos in the Alboran Sea or the Algerian-Provençal basin (Véla et al. 2013). Furthermore, a relatively small area sometimes harbours a relatively high level of species richness, such as the archipelago of El Aouana on the Jijel west coast, classified as an Important Plant Area (Véla and Pavon 2013). Finally, the random sharing of a number of endemic and/or threatened species in common with the nearby mainland makes these islands and archipelagos to belong to a mainland IPA: Rachgoun to the Trara Mounts IPA, Sekia and Dziria to the Cap Ténès IPA, Islet à l’Ail and Islet Sahel to the Gouraya National Park IPA, Islet and Islet Grand Cavallo and Petit Cavallo to the El Aouana coastal IPA (Benhouhou et al. 2018), Pain de Sucre, Fontaine Romaine, Gargamiz, Kef Amor, Akacha and Toughnechet in the Edough Peninsula IPA, Boutribicha, Callisar and Hennaya (Lehnaya) in the El Kala-1 IPA (Hamel et al. 2023).

At this stage, the island of Serigina, which also hosts some endemic plants (Véla 2008), is not part of an existing IPA on the continent. However, the process of identifying IPAs is still incomplete, as the coastline between Skikda, Stora, and Oued Bibi hosts several endemic and/or threatened species, including Lotus drepanocarpus Durieu and especially Anthemis maritima subsp. bolosii Benedí & Molero (Véla 2008; Sakhraoui et al. 2020; present work), which by themselves justify classification as an IPA (Yahi et al. 2012). Therefore, we propose to denominate this new IPA “Corniche de Stora”, which joins the forty or so IPAs already identified in Algeria (Benhouhou et al. 2018; Mostari et al. 2020). This is also the case for Sbiaat Island, where endemic and rare species presumed to be threatened as Anthemis chrysantha J.Gay subsp. chrysantha and Sonchus tenerrimus subsp. amicus (Faure, Maire & Wilczek) Véla (present work) have been recorded, justifying its classification as an IPA according to the criteria of Yahi et al. (2012) (Table 4). It is worth mentioning that this island belongs to a larger natural complex located on the mainland from Cape Figalo to Cape Lindlès. Here, again, it seems important to name a new IPA “from Cap Figalo to Cap Lindlès”, which extends westwards the IPA “the Oran hills” with a much more pronounced relief (Yahi et al. 2012; Benhouhou et al. 2018).

The floristic factor analysis revealed that the most influential floristic attributes in the separation of the studied Algerian islands are total richness and the proportion of zoochorous species.

In accordance with the classic area-species model (Whittaker and Fernández-Palacios 2007), the total floristic richness is strongly correlated with the island area (r = 0.581, p < 0.001). This relationship was also highlighted, for example, by Médail and Vidal (1998), who stipulated that in the western Mediterranean, the surface area of island groups is the main variable involved in the organisation of the richness and composition of the vegetation on these islands, while its relationship with the distance from the coast (site isolation) is less obvious. Benhamiche-Hanifi and Moulaï (2012) highlighted also this relationship for Jijel and Bejaia islands, which is confirmed for the rest of the Algerian coastal islands through our results, which also reveal a negative correlation between the log-transformed island area and the PAR (r = 0.800, p < 0.001).

Regarding the dispersal syndromes of these island floras, the zoochorous and anemochorous syndromes have significant influence in the classification of the islands. The dominance of zoochorous taxa can be explained by the presence of seabirds, in particular colonies of yellow-legged gulls which spread seeds, either actively by eating seeds or fruits (endozoochory), or passively by plumage (epizoochory) (Calvino-Cancela 2011). According to Vidal et al. (1998), these birds could induce a functional link between some small island systems of continental origin close to feeding sites located on the mainland. Their location close to the shoreline has also favoured the settlement and presence of anemochorous taxa on islands which are exposed to onshore winds. The hydrochorous species are nearly present on all investigated islands, unlike the myrmecochorous and autochorous species. Almost all plant species recorded in our study are characterised by entomogamous or anemogamous pollination; favoured by the sea winds as well as the short distances or the weak isolation of almost all our island sites.

The RDA showed that the density of yellow-legged gulls, which is positively correlated with the island area, considerably affects the characteristics of the flora in the islands and islets.

For island flora conservation planning, some of these small islands can be considered “modern refuges” of terrestrial biodiversity from human pressures (Médail 2017, 2022); particularly in the context of Mediterranean ecosystems characterized by a quasi-permanent presence of humans in the various habitats, and notably along the coasts. But these islands themselves are increasingly attracting more curious tourists, and this in proportion to the phenomenon of recent development of seashore tourism (Véla and Pavon 2013; Boutarcha 2019).

The environmental features of these Algerian islands (small area and low slope) make them particularly susceptible to species extinction processes due to sea level rise and genetic drift. Some current estimates, stipulate that a global rise in sea level will need to reach at least 1 m by 2100, and de facto large parts of low-lying island ecosystems are at high risk of being submerged, leading to significant habitat loss (Bellard et al. 2014; Harradine et al. 2015).