Interpolation of agronomic data from plot to field scale: using a clustered versus a spatially randomized block design

Abstract Field trial data are frequently collected at scales different than at which answers are needed. The effects of soil and fertilizer on crop yield is assessed by using usually small, representative and accessible sites. Yield data obtained at plot scale (kg m −2 ) need to be interpolated or extrapolated to obtain yields at field scale (kg ha −1 ), which is the relevant management scale. In soil science it has been recognized that variability of yield data is caused by a combination of fixed and random effects. In this paper we used conventional statistical procedures combined with geostatistics to analyze yield data obtained from soil fertility trials in Cote d'Ivoire. A predictor was formulated which explicitly takes into account main fixed effects as well as spatial dependence. We compared this predictor with a standard application of geostatistical techniques to the measured yield data and with a standard regression procedure. Modelling of fixed effects in combination with kriging was found to be the best predictor. We formulated an alternative to the standard experimental design, the spatially randomized block design, to better accommodate for spatial variability patterns. Distributing the blocks across the catena did not compromise the ability of the design to detect induced treatment effects, but resulted in higher MSE values. For interpolation purposes these higher MSE values are probably more realistic than the MSE values obtained with the classical cluster block design which positions the measurements at small distances.