Using species distribution models to monitor threatened species

Species distribution models (SDMs) are statistical tools that estimate the relationship between species’ occurrence and environmental variables. They are used for many purposes in ecology, conservation, and evolutionary biology, such as to predict the current potential distribution, to evaluate which environmental variables determine the presence of a species, to assess extinction risks, to understand evolutionary niche dynamics, to predict areas that are suitable for reintroduction, or to forecast distribution under climate change. In this study we determined the current distribution of Crocus etruscus Parl., an endemic geophyte present in deciduous woods and pastures of central Italy. Crocus etruscus is included in the National and Global IUCN Red List with the status NT and in Annex IVb of the 92/43/EEC Habitats Directive. Therefore, our goal is to understand the main ecological factors constraining its distribution and use our model to verify new occurrence points in the field to inform its conservation. Species distribution models were built with MaxEnt (version 3.4.3) in R programming language, using 185 occurrences data of C. etruscus taken from Wikiplantbase#Italia and 32 climatic and 22 edaphic variables by Chelsa and Soilgrids, respectively. After running a PCA, we selected the following fourteen variables to include in the models: bulk density of the fine earth fraction, organic carbon density, proportion of sand particles, proportion of clay particles, annual mean temperature, mean diurnal range, temperature seasonality, mean temperature of driest quarter, annual precipitation, precipitation seasonality, precipitation of wettest quarter, elevation, solar radiation in August and in December. All variables were normalized. We used ENMval package to optimize the Maxent model. We set the Regularization Multiplier (RM) parameter from 0.8 to 1.6, and selected 4 feature classes (FC) (L, LQ, LQP, H) to test 20 parameter combinations in total. According to our results, the best model had RM=1.2 and FC=LQ (Fig. 1.a.), with an AUC on the test data of 0.89, reflecting a better ability of the model to discriminate conditions at given occurrence localities compared to those of background localities. Finally, habitat suitability map (Fig. 1.a.) was converted into a binary map (Fig. 1.b.) by selecting as threshold equal training sensitivity and specificity. The model suggests that the distribution area of C. etruscus might be broader than its current range, as suitable areas without reported occurrence data lie southerly along the coast, between Follonica (GR) and Castiglione della Pescaia (GR) and westerly between Monte Cetona (SI) and Monte San Savino (AR). Interestingly, the model highlights the putative presence of C. etruscus on Elba Island where there is the endemic Crocus ilvensis Peruzzi & Carta, suggesting that the two species could share the same environmental conditions.