IntroductionTop

A valuable part of the genetic richness and diversity of forests is held by small populations living at the edges of the natural species range. Those populations, defined as marginal and peripheral populations (MaP populations) will be the first facing the climate change effects and may contain original adaptive traits and genetic variability (Hampe & Petit, 2005). Those will play a key role to study and understand the effects of Global Change on forest ecosystems. Climate change may have different impacts on different populations and expected changes in tree growth will influence the competitive relationships between species. The potential mixture and the choice of species available for plantation or natural regeneration (Cha, 1997; Lindner, 2000) could be modified as well as the frequency of drought stresses or fires (Resco De Dios et al., 2007; Vázquez et al., 2015). Resilience of forest ecosystems and reaction to disturbances are strictly connected to genetic variability and phenotypic plasticity of populations (Ducci, 2015).

It is well known that the future events might influence biomass stocks (Benito-Garzón & Fernández-Manjarrés, 2015), carbon storage and water balance (Vitale et al., 2012) and could force species to migrate from their present geographical range (Parmesan, 2006). Under climate change effects, forest species may have to adapt to new environmental conditions to avoid local extinction or severe genetic erosion (Provan & Maggs, 2012; Hamann & Aitken, 2013). A good prediction of the most likely effects of movement of climatic belts is fundamental to balance future forest management and seed transfers among different ecological regions (Benito-Garzón & Fernández-Manjarrés 2015). Small and isolated populations with low gene flow and low genetic variability could disappear, with a loss of adaptive richness (Schueler et al., 2014) threatened by the speed in the changing environment (Mátýas et al., 2009).

To apply climate change prediction on ecological systems different disciplines are involved. On one side, climatology is fundamental to outline future scenarios. On the other side, ecology and biology are basic to weight and balance species’ response, taking also into account the interaction of biotic versus abiotic factors, especially at the margins of the natural range (Guisan & Zimmermann, 2000) or for planted/introduced species (Isaac-Renton et al., 2014). In such a background, prediction of future impacts on forest ecosystems is an ensemble of climatic scenarios and adaptability of the species that must be considered in a holistic view and tackled under many different aspects (Trivedi et al., 2008) and mainly driven by uncertainties (Araújo et al., 2005; Wang et al., 2012).

To predict such events on forest species many models have been proposed and named as Ecological Niche Models (ENM) and/or Species Distribution Models (SDM) (Guisan & Zimmermann, 2000; Elith & Leathwick, 2009; Warren, 2012; Flower et al., 2013; Mcinerny & Etienne, 2013; Warren, 2013). In those modelling procedures the spatial distribution of forest species is connected to a set of environmental variables that can be used as predictors modelling procedure (Guisan & Thuiller, 2005; Elith & Leathwick, 2009). A vast number of different SDMs (and ENMs) algorithms have been proposed in the literature, often compared with each other in order to assess their power and suitability according to the different nature of data or species distributions (Zaniewski et al., 2002; Liu et al., 2011; Merow et al., 2014). When a SDM (or an ENM) incorporates future climate predictions, future distribution of species can be forecast becoming a very powerful way to study climate change effects on populations (Forester et al., 2013; Brunetti et al., 2014; Isaac-Renton et al., 2014) small parts of species range (Hamann & Aitken, 2013) or provenances (saac-Renton et al., 2014I).

In the Mediterranean region, the Italian peninsula is a well-known hotspot of genetic diversity, consequence of the presence of many ancient glacial refugia (Petit et al., 2003). In this paper, through a SDM approach, the future impacts of climate change on a MAP of Pinus nigra in central Italy (Abruzzo region) were investigated. The aim was to guide the future management processes and to consider and possibly to implement assisted migration actions of this marginal popolation as well as to add knowledge about the more likely events of forest species at the borders of the natural distributions. The current distribution of the species across the region was considered to cover the ecological niche (ecological niche) while just the spatial distribution and shape of the MaP population was investigated. Three modelling techniques were compared and used to create an ensemble model with two future climatic scenarios for central Italy according to two future trajectories from the IPCC AR5, the rcp4.5 and rpc8.5 (IPCC, 2014).

MethodsTop

ResultsTop

Predictors were initially tested for collinearity (Montgomery et al., 2012) and the 5 biovariables with no collinearity problems (Table 1) were added to the soil map and used for building the SDM. The importance of each predictive variable is reported in Table 2. The modelling procedure detected the bio5 (Max Temperature of Warmest Month) as the most important predictor for all models. Climatic data were detected as much more relevant than soil data, especially in MARS model where they were not useful at all.

Table 1. Collinearity test results of the 19 input parameters

Table 2. Predictors’ importance for each algorithm and mean values (range from 0 to 1)

Mean AUC and TSS values of 30 bootstrap runs and of full models for each dataset are reported in Table 3, whereas global means and standard deviations are reported in Table 4. ANOVA performed on TSS values demonstrated the absence of statistical differences between the algorithms. Consequently, the ensemble model was created with all the algorithms. In the current scenario (ABR0) the ensemble model calculated 229,991 hectares of potentially-suitable area, much higher than the present distribution which is 19,185 hectares. In ABR4.5 and ABR8.5 this estimated area decreased very strongly and respectively of -72.1% and -96.5%. According to this prediction, also elevation of suitable envelope was predicted to change. Minimum elevation of Black pine’s populations in Abruzzo was increased from 342 metres a.s.l. of the present potential distribution to 1,530 metres of the rcp8.5 scenario (+347.4%). Mean elevation shifted of approximately +700 metres and maximum elevation reached 2,431 metres (+18.4%). In Tables 5 and 6 the values related to potential suitable area and elevation limits for the three modelled scenarios are reported. A cartographic representation of the projections is shown in Fig. 4. The model correctly predicted the current natural distribution in ABR0, which was drawn as black polygons. With ABR8.5 the situation was completely changed and current distribution was not predicted to be suitable any more. Three different zones were selected by the model on top of the mountains surrounding the Sangro river. The first one was on the higher parts of the Camosciara area (Monte Capraro, Monte Petroso and Monte Tartaro) whereas other two were on the opposite side of the valley (Monte Greco, Monte Marsicano and Monte della Corte).

Table 3. Mean AUC and TSS of 30 bootstrap reperirions

Table 4. Global mean and standard deviation of AUC and TSS values

Table 5. Potential distribution area (ha) in the whole Abruzzo region in the three considered scenarios (number of cells > 0.7)

Table 6. Elevation variance in metres (and percentage referring to ABR0) for each scenario (cell value > 0.7)

Discussion and ConclusionTop

Despite many recent works used RF as unique (or unique-based) algorithm to predict present and future distribution of forest species (Wang et al., 2012; Melini, 2013; Isaac-Renton et al., 2014), also regression-based models (GLM and MARS) performed well in this study. As expected, RF showed higher TSS and AUC and smaller standard deviations. The use of a group of models for a consensus map, with more runs and multiple datasets with biomod2 ensemble weighting method, can consistently improve the prediction of a single model (Marmion et al., 2009; Forester et al., 2013). An ensemble modelling approach can correct biases in calculations, making the prediction more stable than the classical packages of the various algorithms (Liaw & Wiener, 2002). For instance, while RF prediction is often affected by random extraction of trees and variables, the possible overfitting of some models such as GLM and MARS can be reduced.

The current distribution of the Black pine population was predicted to be modified by the considered scenarios. Strong changes were predicted to happen especially with the warmest projection (ABR8.5). The predicted loss of suitable area for the whole Abruzzo region was very high (-95%) and just few mountainous zones were detected as suitable (Fig. 5). However, as the model suggested, Abruzzo’s topographic morphology may play a key role in the conservation of this marginal gene pool. In such case, trees at higher elevations on rocks, growing at around 2,000 metres a.s.l., might increase their reproductive role in a warming climate allowing the species to migrate. In this context, this “new” source of seeds could be a relevant advantage versus competitors such as Beech and/or Oaks which are frequently mixed with Black pine. These hardwood species are not able to migrate at higher elevation without the help of animals. In addition future-suitable lands are actually bare and inhospitable for trees species and, Black pine would probably be more able to colonize these new environments. A similar effect of the elevation ranges was detected in other Italian regions (Attorre et al., 2011; Vacchiano & Motta, 2014).

Our model reported a very high variation of suitable area in Abruzzo predicting higher values for ABR0 than the real situation (+2,000%). Reasons rely on the fact that SDMs works with the occupied ecological niche, compared to the environmental variability. In this view, the spatial distribution of a species cannot match all the locations that have similar conditions. Forest ecosystems are dynamic and complex systems with a mixture of species which interact and compete for natural resources (Ciancio & Nocentini, 2011) and those dynamics can hardly be included in a model. In addition, we must also consider that Black pine is a very plastic species which can grow in a very wide spectrum of areas (Vidakovic, 1974; Isajev et al., 2004; Corona & Nocentini, 2009). Different biotic and a-biotic factors are involved (Pearson & Dawson, 2003). In this view, a specific conservation strategy, based on silvicultural management and monitoring system for the species should be considered in order to observe future development and manage the forest genetic resource properly. If on one side the adaptation strategies are partially considered in this modelling approach, future climate developments are likely to be faster than the migration and adaptation ability of the species (Mátýas et al., 2009). An assisted migration protocol could be taken into account and combined with in-situ adaptive management. A silvicultural approach aimed at increasing the genetic exchange among trees and based on natural regeneration could favour the development of adaptive traits (Brang et al., 2014). This could be achieved with silvicultural interventions which differentiate stand structure and open up the canopy, e.g. small group selection felling (Ciancio et al., 2006). At the same time the establishment of seed orchards and dynamic ex-situ conservation could allow the conservation of the available gene pool.

In the end, predicting the impact of Climate Change on forest species is full of uncertainties. Biological, genetic and ecological skills are fundamental to contribute to these studies, to enforce and validate statistical models and to combine different approaches. Species-specific analysis like genetic diversity, dendrochronology and water-stress resistance could be added to study the phenotypic plasticity of the species (Grivet et al., 2013). In addition, the genetic information about species and their local adaptation must be carefully considered and all the efforts made on local studies, such as those on MaP populations, could have a global impact on the development of a common scientific knowledge base about adaptive processes of forest species.