02046nam a2200181 a 450000100080000000500110000800800410001910000190006024500670007926000390014630000100018550000170019552015560021265000240176865000170179265000280180965000270183715596192000-02-14       1981    bl uuuu  m   00u1 u    #d1 aMENK, J. R. F.  aMathematical classification techniques applied to soil survey.  aOxford: University of Oxfordc1981  a197p.  aPhD. Thesis.  aThis study investigates the feasibility of classifying a very large population on samples conventionally described. It aims at searching for a classification that both minimizes the within-class variation and produces reasonable stability. This study also examines optimal segmentation methods for determining the boundary positions on transects over the area. The study area is in Devon, south-west England, near Ivybridge. The rectangular area, 6x10km, had already been sampled at the intersections of a 100 m square grid. 6060 sites had been visited and about 5830 soil profiles examined by screw auger to 1 m or to impenetrable layer, whichever was encountered first. Small pits had been dug where podzol soils occurred. At each observation site, horizons had been identified and  several soil properties measured. The resultant data contain several types of variables and many missing values. These data had been stored in data files in the computer at Rothamsted Experimental Station. It is shown that the way of comparing conventional soil profiles is to measure every similarity between two profiles from each similarity between two profiles from each similarity between comparable horizons or layers at similar depth. Gower's general similarity coefficient seemed the most suitable measure of likeness for calculating similarity between individuals described by mixed type of variables. It was modified in this study to match horizons or layers for comparing soil profiles. On this basis, a matrix of similarities between profiles was formed.  asoil classification  asoil surveys  aClassificação do Solo  aReconhecimento do Solo