Fjölrit RALA - 15.06.2004, Blaðsíða 22

To confirm this interpretation we performed a discriminant analysis, using as classification
criteria the andic nature (i.e. 5 aluandic, 41 silandic, 23 vitric, 5 organic and 20 non-andic
horizons).
Figure 1. Projection of the first two axis of the PCA

2,5 ■ • X □ □
2,0 • Organic
Aluandic
1,5 ■ Vitric ' • X X o □
1,0 ■ • • X □ □
• □ □
0,5 ■ V. □ □ m □
Non andic ®o • □ □ □ E3
0,0 ■ o / • □ □ □ s o o □
E3 □
-0,5 ' o° • m □ Silandic
-1,0 ■ o oo %• • tya Eq
o° 0 • □ □ E3 □ □ □ □ □
-1,5 ■
2,0 -
-2,0 -1,5 -1,0 -0,5 0,0 0,5 1,0 1,5 2,0 2,5 3,0
Factor 1
Two canonical functions explained 98% of the variance and included: P retention, total C,
pH in NaF, pH in water and Alp as predictor variables. Again, the first canonical function
separated the horizons by their andic/non-andic nature and the second by the role of the
organic matter. The elimination of the organic horizons gave also two significant canonical
functions including P retention, pH in water, Alo and Feo; while the elimination of P retention
resulted in inclusion of pH in water, pH in NaF, Alo, Sio and the ratio Feo/Alo. Thus, if
organic horizons are not considered only variables related to the reactive components are
significant in the separation. These resutls are consistent with the nature of andic soils and
with the criteria required for their characterization.
These results are promising but also indicate that the disgnificance of the analysis could be
improved by 1) enlarging the database of andic volcanic soils, 2) include non-volcanic andic
soils and 3) use other statistical methods such as SEM (structural equation modelling), which
take advantage of the large redundancy among soil the properties.
11