MAPPING ORDINAL DATA IN SOIL SURVEY - A COSTA-RICAN EXAMPLE

In most soil surveys, and particularly those carried out in developing countries, the bulk of recorded soil data is on an ordinal measurement scale. This determines the type of statistics and techniques that can be used for data processing. For example, with ordinal data the mode, median, and range can be used as summary statistics. For describing the spatial structure of ordinal data we propose the spatial-difference-probability function, which is comparable with the semi-variogram. From a soil survey in Costa Rica, three different suitability maps for banana (Musa x paradisiaca L.) were produced according to a qualitative land-evaluation procedure by interpreting: (i) the 1:200 000 soil map, (ii) the 1:50 000 soil map, and (iii) point data. The quality of these three maps was tested by looking at the reliability, relevance, and presentation of information, using 98 test borings. Reliability was characterized in terms of purity and range. The suitability map based on the 1:200 000 soil map was the most reliable one, with an overall purity of 49%. The suitability map produced by interpolating point data was the most relevant one, as defined in terms of the possibility to correctly identify potential locations for banana plantations from the map. Differences in presentation of information were evaluated by comparing the boundary indices of the different maps. The suitability map based on the 1:50 000 soil map was the most readable one. Map choice should be based on a consideration by the user of the different quality criteria.