*4.1. Species-Specific Requirements against a Changing Climate*

The species-specific ecological requirements of forest tree species are one of the main drivers for ecological modelling. While similar output can be obtained with species sharing the same climatic envelope (i.e., silver fir and European beech), different projections are instead calculated for species that are highly differentiated (e.g., European beech and holm oak). Even if just one RCP scenario was used in this study, large differences were found between RCMs and GCMs. Our results underly how the uncertainty on climate change projections have a great impact on spatial model simulation. The use of different types of climatic data (GCMs or RCMs) can lead to very different SDM projections and with potential impacts on SFM decisions [36,71,72]. The use of RCMs with respect to GCMs generally leads to better final climate projection and also to a systematic reduction of bias [73]. This aspect can represent a fair improvement especially for mountainous areas where the use of coarse data can only partially capture the effect of orography [74]. The results we obtained also highlighted the difference in the use of GCMs versus RCMs which are probably optimized scenarios for local areas but very complex and whose calculation is time consuming [61,75,76]. Unfortunately, the use of local data is not still very common and ensemble models are lacking in literature [77]. While several GCMs are sometimes used and then averaged, the use of a single average layer causes the loss of variability with no information on the range of all the potential predictions made by the same SDM procedure. For this reason, an uncertainty assessment should be always mandatory when forecasting climate change impacts. Some papers have also introduced the consensus method to assess the uncertainty in different climate scenarios [78,79], but the use of more GCMs, RCMs, and RCP projections seems to be necessary.

Concerning the mathematical structure of SDM, the importance of the quality of data sources is confirmed as well as its relationship with the uncertainty in species occurrence data and the different statistical technique used to predict the species distribution [45]. Uncertainty in species occurrence data can have a negative effect on the accuracy of a model and any possible correction might bring a potential reduction of the total number of records, removing the uncertain or filtering possible outliers [27]. However, this effect can have different impacts on the SDM according to the modelling technique. Even if MAXENT is the most used in scientific literature and acknowledged as able to provide high accuracy despite the use of occurrence data [31,80], this algorithm was the worst in this study. The reasons might be found in the low number of absences we used (i.e., the background points for MAXENT), probably too few to allow the model to work properly [70]. As a consequence a real and powerful SDM should be based on high-quality data, representative of the phenomena and without any prejudice on the modelling algorithm to be used, with the unbiased comparisons as the unique technique to assess their predictive power [42].

The above-mentioned differences between algorithms, climate projections, target species, LS thresholds to be used for binary transformation etc. (i.e., modelling uncertainties), might heavily impact the operational use of SDM as a decision support system to support strategic sustainable forest management. One of the main uses of SDM is the possibility to identify candidate tree species (genotypes) and provenance types (genotyping) which may be more adapted to future climate conditions in a specific area [9,81]. Provenance selection has the potential to support Assisted Migration strategies (AM) and in-situ or ex-situ conservation efforts to improve the resilience of forest systems [3,82]. While AM represents a possible action for a quick response to climate change threats, this should be realized carefully [10,83]. Such action is probably the most expensive, extreme, and potentially dangerous for ecosystems in case of biased SDM. In fact, despite the advantages attributable to this operation, linked to the avoiding of extinction of species and to supporting economic activity such as timber production, there are many potential disadvantages connected with AM operations that are related to a series of biological risks (the maladaptation or the introduction of invasive species or pests and disease) as well as ethical problems that are connected to the different points of view with respect

to the relationship between nature and human and, therefore, the conflict among anthropocentric and eco-centric positions [84,85]. Consequently, AM must be driven by reliable models, averaging different models and GCMs outputs in a framework of statistical probability. The higher the uncertainty in the modelling steps is, the more dangerous and biased the efforts could be, with the probability of failure which is proportional to the magnitude of disconnection between what is projected and what is likely to occur.
