3.2 Model predictors
To reduce the number of model predictors we performed a correlation analysis between the remote-sensed variables and each sampled variable (tree height and tree density) (Table S2, supporting information). Best correlated predictors were the same for both tree height and density: 1) tree canopy cover (r= 0.62 and r= 0.50, respectively) (Hansen et al., 2013), 2) fraction of photosynthetic active radiation (FPAR) (r= 0.61 and r= 0.49, respectively), 3) leaf area index (LAI) (r= 0.56 and r= 0.50, respectively) (Myneni,  Ranga et al., 2015), 4) mean land surface temperature (r= -0.42 and r= -0.25, respectively) (Hulley et al., 2009, 2012, 2015; Hulley & Hook, 2008, 2009, 2011; NASA JPL, 2014), 5) temperature annual range (r= -0.50 and r= -0.26, respectively), and 6) temperature seasonality (r= -0.47 and r= -0.23, respectively) both from Hijmans, 2005.