The effect of soil variability on the extrapolation of yield trial results at Gachoka- Embu Kenya

A study was conducted on the effects of soil variability on the extrapolation of yields results on a 2 ha plot at Gachoka, Embu. Five variables - phosphorus, nitro¬gen, organic carbon, day content and pH ¬were appraised for three soilsdepths: 0-10 em, 40-50 cm and 90-100 cm. The nearest neighbour interpolation procedure was used in which observations are assigned weight in proportion to a function of the inverse distance between the interpolation location and the observation locations. The Integrated Land and Water Management Information System (ILWIS) computer pro¬gramme was used to develop distribution maps of the selected soil properties for the threee sampling depths, while the statisfi¬cal analysis was performed via the Geostatisfical Environmental Assessment Software (GEO EAS). Although the spatial dIstribution of phos¬phorus varies in the surface horizon (0-10 em), more than half of the surface area of the geostatistical grid plot belongs to one single P class (7-10 ppm).In the subsoil, the distribution pattern is relatively homoge¬neOllS and comparable to the spatial distri¬bution of nitrogen. The phosphorus content decreases with increasing soil depth. In view of the former, it is unlikely that the spatial distribution of phosphorus could cause substantially large yield variations. The spatial distribution of organic carbon is fairly homogeneous in the surface horizon (0-10 em) as well as in the deeper part of the profile (90-100 cm). This variable does not constrain significantly, extrapolation of trial result data. Similarly, the distribution of nitrogen is horizontally and vertically homogeneous. It is therefore possible to extrapolate trial results data to larger areas. The same conclusion can be formulated about the clay content which is fairly uni¬form through space.In contrast, the pH shows significant varia. tions in all the three soil horizons, and this could cause some limitations to the extrap¬olation of point data results to larger areas.