Spatial variation of soil and plant properties and its effects on the statistical design of a field experiment

Abstract

The standard procedure in field experimentation is to use a randomized complete block (RCB) design, striving for the most homogeneous conditions as possible among plots in the same block. However, geostatistical concepts dictate that a spatial dependence exists for observations of a particular property, where closely spaced are more similar than those taken at a greater distance. The present study was conducted on an 18 ha lucerne (Medicago sativa) stand in which a 113 m X 145 m experimental area was demarcated. To determine spatial characteristics of soil and plant properties, 48 sampling points were selected using a 20 m square grid with an additional 75 points on a 2.5 m grid at five random node points. A RCB design trial was superimposed on the geostatistical grid design and consisted of seven pseudo (i.e. non-existent) treatments, replicated four times. Soil and plant samples were taken in June 2001 at all sampling points and plots and analyzed for various properties, including green biomass yield. Analysis of variance of the RCB design revealed statistically non-significant differences among the pseudo treatments, as expected, for various soil and plant properties, including yield. Although large yield differences (2.1 – 5.3 t ha-1) were observed, the conclusion could be made that the experimental field was homogeneous enough to lay out a RCB design experiment. However, it was found that the estimate map of soil pH(H2O) indicated a spatial dependence. The question was posed that if spatial variation in soil pH(H2O) had been considered, would it have had any effect on the results of this field experiment, for example, in terms of yield? Scrutiny of the latter variability revealed that the standard RCB designs did not provide homogeneous blocks with respect to soil variability. The redesign of the experiment where all plots were randomly allocated to treatments and replications, led to significant differences obtained for plant and soil properties as a function of the pseudo treatments. Analysis of co-variance was then applied to eliminate variability between plots and the resultant pseudo treatments showed no significant differences. From this study it became evident that spatial variability of soil and plant properties can jeopardize the results of a standard RCB designed field experiment and should be taken cognizance of.