^{1}

^{1}

^{2}

^{3}

^{1}

The authors have declared that no competing interests exist.

Due to the impelling urgency of plant conservation and the increasing availability of high resolution spatially interpolated (e.g. climate variables) and categorical data (e.g. land cover and vegetation type), many recent studies have examined relationships among plant species distributions and a diversified set of explanatory factors; nevertheless, global and regional patterns of endemic plant richness remain in many cases unexplained. One such pattern is the 294 endemic vascular plant

The question of how plant diversity is distributed on Earth has long fascinated and inspired biogeographers and ecologists. Due to the urgency of plant conservation and an increase in the amount of high resolution data available, many studies have explored how plant species’ richness results from interactions among topography, geology, climate and anthropogenic factors [

Since endemic plants are frequently threatened, they constitute a pivotal group for conservation [

The five Mediterranean climate regions have been the site of many studies about endemic plant richness [

This research is a representative case study of Mediterranean endemic plants. Indeed, Sardinia is the second largest island of the Mediterranean Basin and it is considered an important centre of plant endemisms [

Sardinia (Italy) and its ca. 399 satellite minor islands are located in the central part of the western Mediterranean Basin and cover a surface area of around 24,090 km^{2}. In the Mediterranean biogeographic region, Sardinia is particularly related to Corsica and the Tuscan Archipelago; together these three areas constitute an independent biogeographical province [

Sardinia is mainly mountainous (

Maps on the spatial distribution of (a) elevation, (b) the simplified lithology subdivided into six categories: Quaternary sedimentary outcrops (Q_sedimentary), Tertiary limestone outcrops (T_limestones), Tertiary volcanic outcrops (T_vulc), Mesozoic limestone outcrops (M_limestones), Paleozoic metamorphic outcrops (P_meta) and Paleozoic intrusive outcrops (P_intrusive) and the (c) HII [

There are a total of 2,494

The geodatabase of all EVP was assembled from published literature,

Subsequently, from the 60,301 EVP occurrence records, we built a 1×1 km grid-based matrix for all Sardinian territory to account for three response variables: (1) the richness of exclusive EVP (hereafter, exclusive EVPR), (2) the richness of insular EVP (insular EVPR) and (3) the richness of all EVP (total EVPR). From the initial number of 36,235 cells, our analyses were restricted to only those grid cells with a minimum of one exclusive (2466), insular (34,375) or total EVP (34,603 cells). This also allowed us to reduce problems related to sampling bias, as all cells used were visited by the authors or by other botanists during recent decades. For the 1 km resolution grid map of the three response variables, see the supporting information (

All explanatory variables used for this study were derived from high-resolution free datasets. A total of 16 predictors were subdivided into three groups: topography and geology (five variables), climate (six variables), and human influence (five variables).

We used two variables, elevation and slope, that are strictly associated with topography and three further variables related to geology: number of geological units, number of land units and lithology. Elevation and slope were computed by averaging values from a 10 m resolution Digital Terrain Model (DTM; available at the institutional Sardinian geoportal,

Six bioclimatic variables from the WorldClim database version 1.4 (years 1950–2000) [

We used five variables related to human influence. The first four of them were obtained from the institutional Sardinian geoportal (

Methods of variable reduction to avoid collinearity were carried out following Irl

The EVPR of all groups of endemics were then fitted by using Generalized Linear Models (GLMs) with Poisson error distribution and log-link function. Likelihood ratio tests based forward selection were applied to check for any significant improvement within models, where variables were included if the related p-value was above 0.05 and removed if the related p-value was above 0.10.

Therefore, variance partitioning for GLMs was implemented to assess the overall importance of climate, topography and human influence [^{2}) [^{2} measures how much deviance a given model explains compared to a model with no variables (the null model) [

The Unexplained (U) and the explained variance of each group of explanatory variables (Human influence (Human), Climate, and Topography and geology (Topography)) are shown on the left. Figures on the right display the relative importance of each explanatory variable calculated as the normalised percentage contribution to the adjusted R^{2} for the respective response variable. See

Total EVPR | |||||
---|---|---|---|---|---|

Variables^{a} |
Categories^{b} |
Estimate | z-value | χ2 | P^{c} |

HII | H | -1.71 | -3.13 | 0.2 | |

LU_Ratio | H | -1.56 | -18.02 | 6.7 | |

Roads | H | -4.29 | -11.60 | 136.2 | |

N_Land | T | -5.87 | -2.58 | 330.8 | |

Slope | T | 8.48 | 17.05 | 293.2 | |

N_Geol | T | 1.00 | 0.49 | 9.6 | 0.625 |

Elev | T | 1.95 | 101.06 | 9858.2 | |

Bio7 | C | 1.75 | 9.96 | 98.8 | |

Bio15 | C | -8.50 | -16.52 | 270.2 | |

HII | H | -2.12 | 5.63 | 14.3 | |

Fires | H | -3.51 | -4.67 | 21.3 | |

LU_ratio | H | -4.62 | -2.16 | 4.7 | |

Roads | H | -2.62 | -2.65 | 4.2 | |

N_Land | T | -1.27 | -14.24 | 205.8 | |

N_Geol | T | -3.05 | -1.30 | 1.7 | 0.194 |

Elev | T | 1.68 | 82.59 | 6594.8 | |

Slope | T | 8.35 | 16.10 | 261.5 | |

Bio7 | C | 8.85 | 4.86 | 122.7 | |

Bio15 | C | -6.01 | -11.13 | 23.6 | |

HII | H | -0.04 | -0.21 | 0.04 | 0.832 |

LU_ratio | H | -2.91 | -3.21 | 10.51 | |

Elev | T | 8.83 | 15.57 | 258.76 | |

Slope | T | 6.66 | 4.07 | 16.69 | |

Bio15 | C | 2.67 | -4.71 | 22.23 |

^{a} Variable abbreviations: HII = Human Influence Index; Fires = index of fires occurred among the years 2005–2013; LU_ratio = ratio of 1–2 Land Use first levels (i.e. anthropogenic uses) and the total surface; Roads = kilometres of roads per grid; N_Geol = number of geological units; N_Land = number of land units; Elev = elevation; Bio7 = annual range of temperature; Bio15 = precipitation seasonality.

^{b} H = Human influence; T = Topography and geology; C = Climate

^{c} Significance (in bold for P < 0.05) of the likelihood ratio tests (LRT) was determined using the Chi-Squared (χ2) contribution with 1 degree of freedom

In addition, the percentage of relative importance for each response variable was calculated using hierarchical partitioning of variance, employing the

To investigate the specific relationship between elevation and EVPR and a possible interaction between elevation and area, the analysed region was subdivided into 100 m elevation intervals and the variation in number of 1 km grid cells per each interval was plotted and compared with the variation in local (number of endemic plant

After excluding collinear and weak explanatory predictors, there were eight remaining correlated variables with significant relationships for total EVPR, eight for insular EVPR and four for exclusive EVPR (

All predictors related to human influence, the number of land units, and precipitation seasonality (Bio15), demonstrated a negative correlation with all groups of EVPR. On the contrary, EVPR increased with elevation, slope and annual temperature ranges (Bio7).

In every case, elevation alone accounted for more variance than all other variables together (

Comparisons among variations in exclusive, insular and total EVPR and elevation stressed that the three EVP groups showed similar exponential patterns with the highest local EVPR in cells at the highest elevations (approx. > 1300 m a.s.l.) (

Variations in local EVPR (number of endemic plant

As previously found for other Mediterranean continental islands [

Considering that high elevations comprise smaller areas, high EVPR are also reflecting high EVP concentrations. On the other hand, the area

The increase in endemic plant species’ richness as elevation increases was also found in other Mediterranean contexts [

Species composition, and the richness in the most interesting areas of endemisms (mainly mountainous areas, but also some coastal areas, such as small islets and cliffs), were also related to the ancient traditional land use of ecosystems [

Relationships between EVPR and elevation might be influenced by other important factors, mainly climate, the effects of which might be partly masked by elevation or were not measurable. According to previous researches e.g. [

Local EVPR was significantly influenced by variables, such as the slope and the number of land units, which are often in synergy with elevation [

Although it is common to find a positive effect of habitat diversity on EVPR at regional scale by increasing space available for niche partitioning and speciation and, thus, for species coexistence [

Assuming that richness of endemisms reflects species speciation rate reasonably well [

For the first time, our study provides a general-picture, from the lowest point on the coast up to the highest mountain peaks, about the distribution patterns of all endemic vascular flora of Sardinia. Nonetheless, a large section of variance remains unexplained, mainly because the distribution of EVPR can hardly be related to all possible past and present causes. Since the relationship between EVPR and elevation might be sensitive to the sampling grain, a possible way to improve the knowledge in this field could be to compare analyses and results at different resolutions by considering the same parameters inside both coarser or, if possible, finer grids. Alternatively, specific local studies or, despite their costs, species-specific empirical researches may be the only feasible approach for understanding some specific issues.

(DOCX)

(DOCX)

(DOCX)

(DOCX)

(ZIP)

^{st}century climate change