*2.2. Sources of Data and Environmental and Edaphic Variables*

The data used in this analysis come from the National Forest Inventory developed by the General Directorate of Forestry for the years 2010, 2011, 2014, and 2016 [27], in which 128 parcels of *Eucalyptus dunnii* and 326 parcels of *E. grandis* were measured. The number of trees measured was: 5041, 1200, and 3561 in the first, second, and last inventory, respectively. The number of plots was determined to have a sampling error of less than 10%. The age range measured was from 5 to more than 20 years, the temporary plots were circular with an area of 113, 314, 616, and 1018 m2 (6, 10, 14, and 18 m radius, respectively) and the number of trees in each plot ranged between 5 to 59 individuals. In each of the plots the diameter breast height as measured, the total and commercial height of all trees that exceed a height of 1.3 m and with diameters less than 10 cm and based on these data, survival was calculated. The diameter breast height was measured with a caliper with an accuracy of 1 mm, with two measurements perpendicular per tree and the height was measured with Vertex and Bitterlich's relascope [27]. These occurrence records were spatially filtered to avoid spatial autocorrelation [28] and spatial sampling bias [29,30]. There were generated ten sets of equal number, to occurrence for each species, of randomly distributed pseudo-absences [31] within the study area, with a minimum spatial distance of ~1.5 km, larger than the hypotenuse of the pixel resolution, using the *BIOMOD\_FormatingData* within the biomod2 R package [22]. Additionally, an equivalent random number of absence data were generated covering the area where the species are not present in Uruguay. Those areas were identified corresponding to plots of the National Forest Inventory without the presence of selected species.

A preliminary set of 19 bioclimatic and 48 monthly climatic variables were selected, either for the current situation or for future projections for the specific years 2050 and 2070. These were obtained from the Worldclim database [32] with a resolution of ~1 km (Table 1). A model of global circulation of the atmosphere (GCM), which provides projections of the carbon dioxide concentration in the atmosphere [33] and considers a climate reconstruction (Community Climate System Model, CCSM4), and four representative pathway scenarios (RCP) of different greenhouse concentrations (2.6, 4.5, 6.0, 8.5) were used. We selected the CCSM4 climate change scenario because it is close to the ensemble average of whole GCMs both in terms of temperatures and rainfall and also because have been proof to present the highest accuracy to estimate regional temperature in north-eastern Argentina [34], close by our study area. The chosen values represent the increase in the heat absorbed by the Earth (for the year 2100) according to the concentration of greenhouse gases in each projection, measured in Watts per square meter. In addition, we used 20 edaphic (scale 1:40,000) and topographical variables [35], which were considered as constant for future projections.



Source: http://www.worldclim.org/; # 1–12: January to December.
