Computer-based soil mapping of small areas from sample data II Classification smoothing

Summary Two rectangular areas 1400 m ? 600 m in south central England had been sampled at the intersections of 100 m square grids, and some 20 soil properties measured. Hierarchical numerical classifications gave classes which, when mapped, were very fragmented. Fragmentation was less when the geographic locations of the sampling sites were included in the classification. Linear functions of location and contiguity constraints produce maps that are smoother, i.e. less fragmented, but are unsatisfactory in other aspects. Inverse square and exponential functions of location produce smoother maps that are quite as good as those produced without taking location into account, in that the soil is no more variable within the classes mapped.