Intersite Transfer of Estimated Response Surfaces

Abstract

Transferability of agrotechnology assumes the feasibility of extrapolating a response-input relationship, estimated from experimental sites, to other sites with similar conditions. One specific conjecture is that crop production technology is transferable across sites within a soil family classification. The general approach to evaluating the transfer conjecture involves incorporating into the data analysis the prediction of yields not used in the estimation of the transfer function. A transfer model, using a second-order response surface and measured site-variable information, is formulated, and the yields for each experimental site are predicted from a transfer function estimated from the other sites. The resulting transfer residuals are compared with the ordinary within-site residuals. Based on a sum-of-squares criterion, a prediction test statistic is developed and shown to have a distribution of a ratio of independent quadratic forms. The methodology of transfer residuals is applied to data from the Benchmark Soils Project, where the major objective is to assess the feasibility of transferring agrotechnology among sites having soil of the same taxonomic classification.