based on regression tree method from Po Plain Italy, involving air temperature (atemp), elevation (elv), and B (Boron concentration)
Based on the results above we can draw some key points:
  1. Regression tree is a quantitative classification method providing exact numerical values for each separation;
  2. The method relies mainly in the data distribution given no collinearity occurs in the dataset;
  3. In case of collinearity occurrence, one should drop the method and proceed to another multivariate method in which does not depend on collinearity, in our case is PCA.
Then we shift to PCA. Our model indicates that the first three principal components (PCs) together account for 86.1% of the total variance in the dataset, in which the first PC is 59.3%, second PC is 74.2%, and the third PC is 86,1% of the total variance (Table 4). The concentrations of B, Fe, Na, K, and Li, show high positive loadings (0.29) whereas concentrations of pH, As, and air temperature have low negative loadings (0.16-0.34) for the first PC. In the second PC elevation has high positive loadings (0.35) and the concentration of Ca shows low negative loading (-0,42). The loading values resemble the portion of influence of each variable to a particular sample. Therefore we could build a sample wise distribution plot based on those values as seen in Figure 4.
Table 4 The principal components (PC) extracted from the model