Classification of the Mediterranean lowland to submontane pine forest vegetation

Aim: Vegetation types of Mediterranean thermophilous pine forests dominated by Pinus brutia , Pinus halepensis , Pinus pinaster and Pinus pinea were studied in various areas. However, a comprehensive formal vegetation classification of these forests based on a detailed data analysis has never been developed. Our aim is to provide the first broad-scale classification of these pine forests based on a large data set of vegetation plots. Location: Southern Europe, North Africa, Levant, Anatolia, Crimea and the Caucasus. Methods: We prepared a data set of European and Mediterranean pine forest vegetation plots. We selected 7,277 plots dominated by the cold-sensitive Mediterranean pine species Pinus brutia , Pinus halepensis , Pinus pinaster and Pinus pinea . We classi - fied these plots using TWINSPAN, interpreted the ecologically and biogeographically homogeneous TWINSPAN clusters as alliances, and developed an expert system for automatic vegetation classification at the class, order and alliance levels. Results: We described Pinetea halepensis as a new class for the Mediterranean low-land to submontane pine forests, included in the existing Pinetalia halepensis order, and distinguished 12 alliances of native thermophilous pine forests, including four newly described and three informal groups merging supposedly native stands and old-established plantations. The main gradients in species composition reflect ele-vational vegetation belts and the west–east, and partly north–south, biogeographical differences. Both temperature and precipitation seasonality co-vary with these gradients. Conclusions: We provide the first formal classification at the order and alliance levels for all the Mediterranean thermophilous pine forests based on vegetation-plot data. This classification includes traditional syntaxa, which have been critically revised, and a new class and four new alliances. We also outline a methodological workflow

dense canopy with sparse undergrowth in places with a long absence of fire (Farjon & Filer, 2013). The distribution ranges of Pinus halepensis and Pinus brutia overlap in Greece and on the Aegean islands.
Pinus pinaster thrives in the western Mediterranean Basin (Abad Viñas et al., 2016a). It is a thermophilous species believed to be native to the Iberian Peninsula, southern France including Corsica, western Italy including Sardinia, and northwestern Africa. It also occurs on the Atlantic coast and reaches about 2,000 m a.s.l. in Morocco. This species is well adapted to fire (Fernandes & Rigolot, 2007;Farjon & Filer, 2013).
Pinus pinea is an iconic Mediterranean thermophilous tree, which occurs from the sea level up to the mountains. It has been frequently planted as an ornamental tree and for its edible seeds.
It occurs at high elevations mainly in the eastern Mediterranean, where it naturally regenerates at some sites (Abad Viñas et al., 2016b).
The current distribution of the Mediterranean pines is influenced by the geological history of the Mediterranean Basin and climatic conditions during the Quaternary (Panetsos, 1981), though there were additional influences by humans. Especially the native | 3 of 37 Applied Vegetation Science BONARI et Al. distribution of Pinus pinea, and to a lesser extent of Pinus pinaster, is uncertain because their current distribution is highly influenced by planting, mainly in the western Mediterranean (Mazzoleni, 2004;Bonari et al., 2020). Humans have always taken advantage of the modest ecological requirements of pines and they have used them extensively in plantations for centuries (Bonari et al., 2017), although on the Iberian Peninsula most of the pine plantations were established only in the 20th century. Plantations are easily identified if they occur outside the native distribution range of the dominant pine species, but old plantations in the native range of the pine species may be difficult to distinguish from natural pine forests. Understorey species composition of pine forests varies considerably in response to many factors, including management and disturbances (Farjon & Filer, 2013;Kavgacı et al., 2017;Bonari et al., 2019a).
While some studies tried to clarify the native range of pine forests (Martínez & Montero, 2004;Bonari et al., 2020), others focused on the plantations and their dynamics, for instance, the expansion of Mediterranean pines from plantations into adjacent natural non-forest plant communities (Lavi et al., 2005). Effects of pine plantations on soil, faunal communities, vegetation, biotic and abiotic gradients were also reviewed (Maestre & Cortina, 2004;Gómez-Aparicio et al., 2009).
The most important contributions to the syntaxonomy of the Mediterranean pine forests so far were made by French authors, especially Pierre Quézel and Marcel Barbéro, who performed extensive field surveys in the Mediterranean Basin from the 1970s to the 1990s. They focused mainly on the eastern Mediterranean Basin , including different parts of Anatolia (Quézel & Pamukçuoǧlu, 1973;Akman et al., 1978Akman et al., , 1979Quézel et al., 1980), Syria , Lebanon (Chouchani et al., 1974;Abi-Saleh et al., 1976), Greece  and Cyprus , but also North Africa Barbéro et al., 1981;Quézel et al., 1987Quézel et al., , 1992Benabid, 1988) and France . These contributions have created a backbone for the syntaxonomical scheme of the Mediterranean thermophilous pine forests for a long time, although significant advances have been achieved since then. The most recent comprehensive classification of European vegetation, EuroVegChecklist (Mucina et al., 2016), included the Mediterranean pine forests in the classes of broad-leaved forests, Quercetea ilicis and Quercetea pubescentis, following the established tradition (Barbéro et al., 1974;Rivas-Martínez, 1974;Quézel & Barbéro, 1986;Rivas-Martínez et al., 1986;Brullo et al., 2008).
There are open questions of paramount importance for the Mediterranean pine forests and their management, such as climate change effects, fire risk, or the dynamics of alien plant invasions.
This research agenda for the near future, as well as conservation planning and management, can be significantly supported by a welltested classification scheme for the Mediterranean pine forest types.
Widely conflicting views on the syntaxonomy of the Mediterranean pine forests still exist even after the publication of EuroVegChecklist (Mucina et al., 2016) because the alliances accepted in this checklist have never been tested with a comprehensive set of vegetation-plot data. Moreover, forests dominated by Mediterranean pines were, at least in the past, not identified as independent syntaxa in spite of their distinct physiognomy and their wide distribution across the Mediterranean Basin. Due to the frequent presence of many macchia species, they were relegated into shrubland vegetation units of the order Pistacio lentisci-Rhamnetalia alaterni. For example, Rivas- Martínez et al. (1986) listed Pinus halepensis as a character species of this order. This is not consistent with the treatment of boreal or temperate pine forests which are classified in different classes than the broad-leaved forests and shrublands. This approach was partly inherited from the view of early researchers who considered the pine forests as non-climax vegetation. However, at least Pinus brutia and Pinus halepensis can form pure climax forests in a suitable climate (Feinbrun, 1959;Quézel, 2000;Boydak, 2004;Bonari et al., 2020). Another explanation lies in the fact that the native distribution of some pine species is contentious (see e.g. Martínez & Montero, 2004), and some of the extant pine forests may have originated as ancient plantations. This may be the cause for the reluctance of phytosociologists to describe syntaxa based on dominant species of uncertain origin. With increasing knowledge of the distribution of both species and communities, new syntaxonomical units of Mediterranean thermophilous pine forests were described in recent years (e.g. Pérez Latorre et al., 1998;Mucina et al., 2009;Biondi et al., 2014;Biondi & Vagge, 2015;Mucina et al., 2016;Pesaresi et al., 2017).
Currently, interest in vegetation classification and its applications is growing . The introduction of new numerical methods and formal classification approaches (De Cáceres et al., 2015)  contributed to overcoming the criticism of subjectivity of the traditional Braun-Blanquet method of vegetation classification (Braun-Blanquet, 1932). This has also paved the way for synthetic international vegetation classification studies on the European scale (e.g. Douda et al., 2016;Peterka et al., 2017;Willner et al., 2017a;Marcenò et al., 2018Marcenò et al., , 2019Landucci et al., 2020). In our case, data from the European Vegetation Archive (EVA; Chytrý et al., 2016) and from the specialized CircumMed Pine Forest Vegetation Database (Bonari et al., 2019b) made it possible to perform a detailed analysis of the Mediterranean pine forests and to accomplish the revision of their classification.
Our aim is to characterize the general diversity of pine forests in the Mediterranean Basin by providing the first comprehensive and internally consistent international classification consensus for the Mediterranean thermophilous low-elevation pine forest types at the alliance level across the Mediterranean Basin and the Black Sea region, based on an analysis of vegetation-plot data.

| Study area
The study area is the Mediterranean Basin and adjacent areas, broadly corresponding to the oceanic Mediterranean bioclimates as defined and mapped by Rivas- Martínez and Rivas Sáenz (2019) for Eurasia and North Africa. It stretches from the Atlantic coasts of Portugal to easternmost Anatolia, measuring approximately 4,300 km along its broadest longitudinal extent (9° W-42° E), and from the Caucasus to Palestine to southern Morocco, extending approximately 1,300 km along its broadest latitudinal extent (48° N-30° N). We considered all the countries bordering the Mediterranean Sea, as well as Portugal, Crimea, the Caucasus and the Euxinian region fringing the southern coast of the Black Sea. The latter three territories were included because of the disjunct native occurrence of Pinus brutia. In the northern part of the range of these forests, orographic features of high mountain ranges protect them from the effects of northerly winds. In addition, the proximity to the Black Sea raises air moisture and precipitation, contrasting with the arid and more continental climates of the surrounding areas. This causes the extension of the distribution range of Pinus brutia forests and many Mediterranean species, which reach as far north as Crimea and the foothills of the Great Caucasus (Didukh, 1992).
The physical-geographic complexity of the Mediterranean Basin needs to be taken into account when dealing with biological communities. Firstly, the Mediterranean Basin encompasses a high number of bedrock types. Limestone is by far the most common, while areas with acidic bedrock are scattered, although locally abundant. Ultramafic rock patches are also present. Bedrock diversity translates into soil diversity (Blondel et al., 2010) and thus into vegetation diversity.
Secondly, the specific Mediterranean climate is characterized by mild, wet winters and warm, dry summers. Temperatures generally increase from north to south. Mean temperatures of the summer months exceed 22 °C but are above 30 °C in some areas. Summers are characterized by the lack of rain, which in combination with high temperatures leads to marked seasonal aridity. The limited occurrence of winter frost is essential for plants. The total annual precipitation is spatially highly variable, ranging from less than 200 mm in North Africa to 2,000 mm in some northern mountainous areas (Lionello, 2012;Rundel et al., 2016).
Thirdly, numerous mountain ranges around the Mediterranean Basin show distinct elevational vegetation belts (Ozenda, 1975;Quézel, 1979;Rivas-Martínez, 1981;Blondel et al., 2010). Different pine species tend to occur at different elevations, although with some overlaps (e.g. Carrión et al., 2000). This allows a clear ecological distinction between two major groups of tree pines in the

| Data set and its standardization
The workflow of this study is summarized in Figure 1. We requested vegetation plots (phytosociological relevés) from EVA (Chytrý et al., 2016)  manuals and taxon concepts. We used the SynBioSys Taxon Database in TURBOVEG 3, which matches the taxon concepts and unifies the taxon names used in different databases included in EVA (Chytrý et al., 2016). Subsequently, we adjusted the taxonomy and nomenclature to the Euro+Med PlantBase (Euro+Med, 2016-2020. The few taxa not included in Euro+Med were named according to the SynBioSys Taxon Database or using the original names given in the source publications or in individual EVA databases. The taxa recorded with different taxonomic resolution were merged into aggregates (e.g. Achillea millefolium aggr., Centaurea alba aggr., Draba verna aggr., Galium mollugo aggr.).
Pines were considered at the species level because subspecies were not always identified in the data set. Also, especially for Pinus pinaster, there is no taxonomic agreement among authors about its subspecies.
Further, we reduced the noise and inconsistencies in the data as follows: (1) bryophytes, lichens and algae were excluded, because they were present only in a subset of vegetation plots; (2) infraspecific taxa were merged into species; (3) species with less than five occurrences in the data set were deleted; (4) tree and shrub species recorded in the herb layer or marked as seedlings or juveniles were deleted; (5) records of the same species in different layers were merged into a single layer; (6) vegetation plots with a size <50 m 2 or >1,000 m 2 were excluded, but plots without size information were retained assuming that most of them were within this size range.
These steps created a data set of 60,735 vegetation plots. The data cleaning was done using the JUICE program v. 7.1 (Tichý, 2002).
To test the differentiation between the Mediterranean thermophilous and non-thermophilous pine forests, and between Mediterranean pine forests and Mediterranean oak forests, we performed an unsupervised classification of the whole data set using TWINSPAN (Hill, 1979; parameters: three pseudospecies cut levels of species percentage cover: 0%, 10%, 25%; minimum group size for division: 10 plots) on a subset of 5,000 plots that were randomly chosen to meet the technical limit of the number of plots that the TWINSPAN program could process. The result is shown in has a transitional distribution between these two groups, but it is more mountainous than the four thermophilous pines. The analysis gives support to the separation in the first division between the Mediterranean thermophilous pine forests and the other pine forests, but not to the separation between Mediterranean pine vs. oak forests (Tables 1 and 2). The floristic criterion used by TWINSPAN does not support this division, which can nevertheless F I G U R E 1 Workflow adopted in this study showing the steps from data set creation to the results (underlined), including vegetation types, characteristic species combination, maps, boxplots and elevational-density graphs. EVA = European Vegetation Archive; CircumMed database = CircumMed Pine Forest Vegetation Database; Ordination = DCA ordination superimposed with climatic variables; Plots = vegetation plots. The extraction of 5,000 random plots from the clean data set for the distinction of low-elevation pine forests from the other pine forests (see paragraph 2.2) and the classification of plots with the EUNIS expert system for the most frequent species of Quercetea ilicis and Pinetea halepensis (see paragraphs 2.4 and 3.1) are not shown in the workflow BONARI et Al.
be based on the stand physiognomy (dominance of conifers vs. broad-leaved evergreen trees).

| Mediterranean thermophilous low-elevation pine-forest data set and resampling
As the TWINSPAN classification showed that the vegetation of forests dominated by the four low-to mid-elevation Mediterranean pines (Pinus brutia, Pinus halepensis, Pinus pinaster and Pinus pinea) differs from the mountain and temperate forests dominated by other pine species (paragraph 2.2), we analysed these forests separately (hereafter for short referred to as Mediterranean pine forests). From the total data set of 60,735 vegetation plots, we selected those in which the total cover of the four Mediterranean pine species was greater than or equal to 15% and exceeded the total cover of the other trees. Where information was available, we excluded vegetation plots sampled in recent plantations located clearly out of the alleged native distribution range of the dominant pine species, while we retained those from putative old-established plantations. Note that it is often not possible to separate native stands from old plantations, especially in the Mediterranean Basin where humans have been changing the landscape for millennia. Delineating what is natural and what is not is even more complicated when working with large vegetation-plot databases, in which more detailed information on individual plots is often missing. The selection resulted in a data set of 7,277 Mediterranean pine forest plots ( Figure 2). The contributions from individual databases are reported in Appendix S1.
At this stage, we removed 381 plots with no coordinates.
Vegetation plots with available coordinates (n = 6,896; Figure 2) were assigned to cells of a geographic grid of 0.6 longitudinal by 0.45 latitudinal minutes, i.e. approximately 50 km × 50 km in the central part of the study area. Subsequently, we performed a geographical resampling in order to overcome the bias due to uneven sampling density across the study area (Knollová et al., 2005). We resampled cells with more than 10 plots per grid cell. This operation removed 650 vegetation plots. In the grid cells that contained more than ten plots, we applied the Heterogeneity-Constrained Random (HCR) resampling algorithm (Lengyel et al., 2011) calculated with Bray-Curtis dissimilarity in plot species composition. This procedure guaranteed that the resampled data set contained, within each cell, plots that were representative of the variation in species composition within that cell. This operation removed 200 vegetation plots. The final data set was a matrix of 6,046 plots and 3,190 taxa (hereafter called "resampled data set" and "Resampled data set 1" in Figure 1).

| Classification and determination of diagnostic species
First, unsupervised divisive classification of the resampled data set was performed using TWINSPAN (Hill, 1979;parameters: three pseudospecies cut levels of species percentage cover: 0%, 10%, 25%; minimum group size for division: 10 plots). Four division levels were used, resulting in 16 clusters. This operation allowed us to understand the coarse patterns of floristic similarity within our data set. With a few exceptions, each cluster contained plots dominated by a single pine species. When no ecologically or biogeographically interpretable dissimilarities in species composition between clusters were found, these clusters were merged. We also interpreted all the clusters syntaxonomically, comparing their floristic, ecological and biogeographical characteristics with the literature. The aim was to identify previously described alliances in our TWINSPAN groups. When the analysis supported the concepts proposed in the literature, we accepted those concepts, meaning that we took a conservative approach. When establishment of a new vegetation unit appeared to be necessary, we considered not TA B L E 1 Shortened synoptic table showing the result of a TWINSPAN classification into two groups of a random selection of 5,000 plots from the initial data set of the Mediterranean pine forest and their related forest types including evergreen oak forests; the numbers in columns 1 and 2 are percentage constancies and points represent species absence; the species shown include the pine species and five other species with the highest value of the phi coefficient (Φ) for one of the two groups; grey shading represents species with Φ > 0.2, Constancy Ratio (CR) > 1.5 and p < 0.05 (based on Fisher's exact test)

Group
No. of plots 1 2 only floristical but also ecological and biogeographical differences from the already established units. We also accepted two types (see paragraphs 3.1.3 and 3.1.4) that did not appear as distinct clusters in the TWINSPAN classification, given the scarcity of plots of these types in the database. We defined them by means of the expert system only. All the analyses were performed in JUICE v. 7.1 (Tichý, 2002). Phytosociological nomenclature is in agreement with the fourth edition of the International Code of Phytosociological Nomenclature (ICPN; Theurillat et al., 2021).
Formal definitions of syntaxa provide reproducible and unambiguous classification (e.g. Chytrý et al., 2020). We prepared formal definitions of the interpreted alliances and informal vegetation types based on the concept of functional species groups (Landucci et al., 2015;Tichý et al., 2019) linked by the logical operators AND, OR and NOT as proposed by Bruelheide (1997). Diagnostic species, determined based on the calculation of the phi coefficient of association (Φ), were calculated for the TWINSPAN-based clusters and used to create the functional species groups and discriminating species groups to be used in the formal definitions. Some of these groups were improved by adding a few species on the basis of expert knowledge. The phi coefficient of association was used as a fidelity measure and calculated for equalized size of clusters following Tichý and Chytrý (2006). We included formal definitions into a classification expert system that is available as TXT file (Appendix S2; for acronyms of vegetation types see paragraphs 3.1.1-3.1.15) and can be run in JUICE v. 7.1 (Tichý, 2002), TURBOVEG 3 (Hennekens, 2015) or R (Bruelheide et al., https://git.loe.auf.uni-rosto ck.de/misc/ESy) We determined diagnostic species of individual alliances based on the data set resampled within grid cells defined as above, but this time nested within alliances ("Resampled data set 2" in Figure 1), meaning that unlike in the "Resampled data set 1," where the geographical resampling was applied to the whole matrix, here this operation was done within the defined alliances to produce reliable diagnostic species. For each alliance or informal group, we resampled cells with more than 10 plots per grid cell. We defined diagnostic species for a particular vegetation type as species with Φ ≥ 0.2, Fisher's exact test p value of the probability of the given concentration of species occurrences within the cluster < 0.05 and Constancy Ratio > 1.5. Constancy Ratio is the ratio between species constancy (relative frequency) in the cluster for which the species has the highest constancy and the maximum constancy recorded in any other cluster (Willner et al., 2017b). We defined constant species as those with relative frequency > 20% and dominant species as those occurring in at least 5% of plots with a cover > 15%.
Based on "Resampled data set 2," we also prepared the ordination diagram, the elevational-density graph and the boxplots for the recognized alliances and informal groups.

To assess differences in species composition between
Mediterranean pine forests and Mediterranean broad-leaved forests, we extracted 1,534 plots classified as "T3A Mediterranean lowland to submontane Pinus forest" and 2,826 vegetation plots as "T21 Mediterranean evergreen Quercus forest" from the EVA database classified by the EUNIS Habitat Classification expert system (EUNIS-ESy v. 2020-06-08; Chytrý et al., 2020). These two habitat types correspond to the classes Pinetea halepensis and Quercetea ilicis, respectively. We identified the species with the highest frequency and calculated their phi coefficient of association for these two habitat types.
All the procedures described in this section were performed using JUICE v. 7.1 (Tichý, 2002).

| Ordination
To relate the differentiation of the accepted alliances to climate, DCA ordination (Hill & Gauch, 1980) of plots was computed with log-transformed percentage covers of species using the vegan package (v. 2.5-6; Oksanen et al., 2019) in R (v. 3.6.1; R Core Team, 2019).
Individual plot coordinates were overlaid with the CHELSA Bioclim data set v. 1.2 (Karger et al., 2017) using the "envfit" function of the vegan package. Climatic data consist of a downscaled model output with temperature and precipitation estimates at a horizontal resolution of 30 arc-seconds (Karger & Zimmermann, 2019). Correlations between 19 climatic variables were calculated using the Spearman correlation coefficient (Sokal & Rohlf, 1995) to reduce the number of available variables. We retained only those variables that were most clearly interpretable from an ecological point of view: mean annual temperature, temperature seasonality (standard deviation of the monthly mean temperatures), annual precipitation and precipitation seasonality (standard deviation of the monthly precipitation estimates expressed as a percentage of the annual mean). The four climatic variables were extracted from vegetation plots with help of the raster package (v. 3.1-5; Hijmans, 2020) using the bilinear method. Apart from the DCA, we displayed the climatic variables in boxplots for each accepted alliance and informal group.

| RE SULTS
We interpreted TWINSPAN clusters mainly at the fourth hierarchical  Figure 3. Also, it is worth mentioning that although many species of Quercetea pubescentis are present in the plots from the coastal areas of the northern Black Sea, TWINSPAN did not separate these plots, most likely due to their very low proportion in the data set. Therefore, these two small clusters (Jasmino fruticantis-Juniperion excelsae and Campanulo sibiricae-Pinion brutiae) represented by a few plots were separated in the expert system. However, most of the TWINSPAN clusters were accepted, either stand-alone or merged, as alliances or informal groups. When a given cluster was split in more than one accepted alliance/informal group, we used the expression "pro parte" ("p.p.").
In contrast, we used the symbol "+" when we merged two clusters.
The diagnostic, constant and dominant species for each accepted cluster after TWINSPAN classification are shown in Appendix S3.
For completeness, we also report the two clusters from Crimea and the Great Caucasus foothills in this Appendix.

| Vegetation types
We classified vegetation plots using the newly created classification expert system for the low-elevation Mediterranean pine forests.
We also defined within the expert system the formulas for Crimean and Caucasian Pinus brutia forests. The expert system included 15 logical definitions of accepted alliances and other vegetation types.
We applied this expert system to the non-resampled data set. The distribution of the plots classified as 12 accepted alliances and three informal groups is shown in Figure 4, along with the supposedly native distribution of the dominant pine species. Shortened lists of diagnostic species are shown in Table 2 Table 3. We include these forest in a new class and a previously       Co-occurring with Pinus pinaster, Pinus pinea macroremains and charcoals are more frequent in the region to the south of Lisbon and have been dated as far back as 6,300-6,400 14 C years BP (Carrión Marco, 2005). Old-established plantations in these coastal environments are indistinguishable from naturally established communities based on their floristic composition. We hypothesize that for long these communities have been shaped by the effect of the cold water of the Atlantic Ocean, which influences local temperature and summer fogginess. Furthermore, strong sea currents and powerful storms support sand deposition, which extends far inland. However, inland plantations, even if old-established, lack the above-mentioned taxa and cannot be considered a part of this alliance. Although Pinus pinaster (and possibly Pinus pinea) was common in inland communities in pre-Holocene times, it declined during the Holocene, being progressively replaced by other Mediterranean species (Figueiral, 1995).  of Pinus pinea on sandstone in the Provence are likely native (Quézel, 1979), as well as those at one site in Sardinia (Arrigoni, 1967). However, the areas currently occupied by Pinus pinea have been artificially extended (and often heavily managed) in recent times. The structure and floristic composition of these forests is highly influenced by management and human impact (Bonari et al., 2019a

| Climatic and elevational patterns
The individual alliances of Mediterranean pine forests mostly occupy distinct elevational ranges ( Figure 6) and are related to different climatic features (Figures 7 and 8). as Pistacio lentisci-Pinion halepensis and Thymo vulgaris-Pinion halepensis, but differs in terms of precipitation seasonality (Figures 6 and 7b).
The Pistacio lentisci-Pinion halepensis is typical of the thermomediterranean belt with a warmer climate and more seasonal precipitation, while Thymo vulgaris-Pinion halepensis occurs mainly in the mesomediterranean belt (Figures 6 and 7b). Rosmarino eriocalycis-Pinion halepensis occurs mainly in the meso-to supramediterranean belts, with a high temperature seasonality (Figures 6 and 7d).

| Mediterranean and non-Mediterranean pine forest alliances
We propose some changes in the system of alliances published in EuroVegChecklist (Mucina et al., 2016) for Europe, but also for North Africa, for which we identified a new alliance of pine (pre-)forests.
Following the physiognomic classification approach at the class and order level, we assigned the alliances of the vegetation dominated by comprises Pinus brutia-dominated forests on the south-facing slopes of the western Great Caucasus above the Black Sea near Novorossiysk (Litvinskaya & Postarnak, 2002;Mucina et al., 2016). The species composition of these forests is close to that of the deciduous forests of the alliance Carpino orientalis-Quercion pubescentis and they were classified to the syntaxa of deciduous thermophilous oak forests in EuroVegChecklist: Quercetalia pubescenti-petraeae and Quercetea pubescentis (but see also Didukh, 1996). Although the Pinus brutia forests in southern Crimea (Jasmino fruticantis-Juniperion excelsae) also contain several species of deciduous oak forests, they harbour more Mediterranean species and structural features than their counterparts in the western Caucasus.
Further studies are needed to clarify the position of the latter forests.
In particular, they will need to be compared with forests of Erico-Pinetea, Brachypodio pinnati-Betuletea pendulae and Quercetea pubescentis.

| Pinetea halepensis: a new class of the Mediterranean thermophilous pine forests
The current European vegetation classification (Mucina et al., 2016) puts a strong emphasis on the physiognomy of the dominant layer in the definitions of vegetation classes. For example, it separates the class of temperate broad-leaved acidophilous forests (Quercetea robori-petraeae) from that of boreal to temperate coniferous forests (Vaccinio-Piceetea) in spite of considerable overlap in species composition, especially in Central European lowland oak and pine forests (Heinken, 2008;Leuschner & Ellenberg, 2017). Similarly, also the non-Mediterranean southern European deciduous oak and pine forests are separated at the class level (Quercetea pubescentis vs. Erico-Pinetea). In this context, the inclusion of the Mediterranean sclerophyllous oak and pine forests in a single class Quercetea ilicis, as proposed by Mucina et al. (2016), is inconsistent, hard to convey to practitioners and difficult to apply in remote sensing of vegetation and land-cover classifications. It also has no clear links to the broadly used systems of habitats or forest types, which usually in the first place make a distinction between broadleaved and coniferous forests (Barbati et al., 2006;Chytrý et al., 2020).
With this in mind, we establish here a new class named Pinetea halepensis to accommodate the Mediterranean thermophilous pine forests addressed in this paper. This class corresponds to the EUNIS habitat type "T3A Mediterranean lowland to submontane Pinus forest," and partly also to "N1G Mediterranean coniferous coastal dune forest" . The new syntaxonomic solution, uniting all of these pine forests in one class, is justified especially by the structural and physiognomic criteria. Also ecologically, natural pine forests are united by their occurrence in either climatically or edaphically extreme environments, such as the most exposed, warm and dry rocky slopes, often on ultramafic bedrocks, marls, dolomites or limestones. This new concept is well supported by the comparative analysis of the phi coefficients for the most frequent species of the classes Pinetea halepensis and Quercetea ilicis (Table 3).  Quézel, 1986 (Mateus & Queiroz, 1993;García-Amorena et al., 2007), but also charcoal remains dating from 33,000 years BP were found in the Lisbon region, making up 93% of all the remains present (Figueiral, 1993 (Bonari et al., 2017). Although most plantations were established in the 20th century (especially on the Iberian Peninsula), in many cases, it is challenging to trace whether or not a pine forest is natural.
Our classification includes informal vegetation types comprising old-established plantations of native pine species, in which natural species composition can develop in the understorey (Bonari et al., 2017(Bonari et al., , 2019a(Bonari et al., , 2020. This is in contrast to what happens in the plantations of most non-native trees (e.g. Eucalyptus spp.   and they are considered valuable for nature conservation (Bonari et al., 2018).

| Data limitations and recommendations for future vegetation surveys
We have laid down a classification which does not pretend to be perfect. We are aware of the fact that some areas were underrepresented in our analyses because of the lower density of plots, meaning that we might have overlooked some vegetation types or some diagnostic species. Nevertheless, a clear advantage of our classification is that it is formally described and reproducible. Also, by covering the full distribution range of the studied dominant pine species, we encompass the full species pool of these forests, so that the alliances can acquire a relevant biogeographical meaning. The expert system (Appendix S2) contains groups of species and decision rules that enable identification of each of the pine forest alliances recognized here.
Some pine-dominated vegetation plots remained unclassified after running the expert system. For example, a considerable proportion of Pinus halepensis-dominated plots, equally distributed around the Mediterranean Basin, remained unclassified at the alliance level, although they were correctly classified at the class and order level.
Because of the open-canopy structure of pine forests, which allows the occurrence of species from various habitats, these plots contained a mixture of species with different ecology. Nevertheless, our expert system showed that there is a large portion of plots with Pinus halepensis that can be classified. The unclassified plots are more or less equally distributed across the study area, which points to local-scale effects (including disturbance such as fire, forestry management or trampling) that make the classification of Mediterranean pine forests challenging.
Further, some areas in our data set are represented by very species-poor plots, in some cases with one to three species only (e.g. in the Levant). Such plots are problematic because they were perhaps sampled in very disturbed areas, but sometimes they were the only data available from a broader area. Disturbances We suggest that with the increasing availability of large vegetation-plot databases and detailed revisions of vegetation classification, broadly conceived geographically defined alliances will be delineated more often than in the past.
Other Mediterranean conifer forests dominated by Juniperus, Cupressus, or Tetraclinis share some structural aspects with pine forests (e.g. relatively open canopies, litter decomposition), but in general, have not been managed so heavily. Some of them reach heights much lower than the pines, but others can be comparable in height when the forest is undisturbed. Their right position in the syntaxonomical scheme of Mediterranean forests should be revised through a large-scale analysis.
Collecting new data in scarcely surveyed areas is needed in the future. In particular, for Pinus brutia, we miss data from the easternmost limit of its distribution (Azerbaijan, Armenia, Iran and Iraq) and also from Israel. For Pinus halepensis, we miss data from northern Libya and Albania. For Pinus pinaster, we miss mainly data from North African countries. For Pinus pinea, we miss data from Southern Balkans and some Mediterranean islands. These new data could provide evidence for recognizing new syntaxa or reinterpreting and redefining the earlier proposed vegetation units.
Despite the Food and Agriculture Organization (FAO) and most of the national forest inventories usually using 20% canopy cover as a threshold for forest, we used a cover of 15%. We advocate that the traditional Braun-Blanquet cover value 2, including covers of 5%-25%, is relatively broad for our purposes and does not ensure by itself that a given plot belongs to a "forest" or shrubland with a few pines. Therefore, the decision to use 15% represents a compromise for not excluding too many pine (pre-)forests plots with an open canopy. For such large-scale analyses, old sampled plots are also crucial, and we used them. However, for sampling new plots, we recommend that at least a cover value of 3 should apply when selecting plot areas for pine forests. Another related problem is that in many old plots the growth form (either shrub or tree) is not indicated for woody plants. Therefore, we also recommend an indication of the height of the strata to recognize the forest structure properly and to evaluate whether or not pines are the dominant trees.

| CON CLUS IONS
We propose a new syntaxonomical scheme for the low-elevation Mediterranean pine forests (Table 4)  In comparison with EuroVegChecklist (Mucina et al., 2016), this study has enriched the syntaxonomical system of Europe by four newly recognized alliances (Coremato albi-Pinion pinastri, Lavandulo pedunculatae-Pinion pinastri, Styraco officinalis-Pinion brutiae) and one newly recognized alliance for North Africa (Rosmarino eriocalycis-Pinion halepensis). One previously invalidly described alliance was validated (Jasmino-Juniperion excelsae). In contrast, a previously described alliance (Rosmarino officinalis-Pinion halepensis) was not supported by the analysis of a large data set. The alliances Alkanno
The results of this study provide information on the compositional and distributional patterns of Mediterranean thermophilous pine forests, offering a list of statistically derived combinations of diagnostic species for the major eco-geographical vegetation units.
Further, the workflow adopted in this study, but also its pitfalls and limitations, might be useful as a pathway for similar broad-scale veg- we thank all the botanists and vegetation ecologists who collected the pine forest data in the field, thus making this synthesis possible.

AUTH O R CO NTR I B UTI O N S
GB and MC conceived the idea; GB, IK and MC developed the database; GB conducted the analysis, under the supervision of MC; SMH and LT provided support for data processing in TURBOVEG 3 and JUICE; JD helped with the interpretation of plant names; KC prepared maps and graphs; GB led the writing, with substantial inputs from MC; all the co-authors participated in discussions and syntaxonomic interpretations.