Implications of choosing different interpolation methods: A case study for soil test phosphorus

Abstract

Many farmers have their fields grid soil sampled to plan for variable rate P fertilizer application. Grid soil samples are often interpolated to create fertilizer application maps. However, most farmers and other practitioners do not compare interpolation methods. The objective of this study was to evaluate the performance of different grid soil sampling interpolation methods on P fertilizer prescription maps. Grid soil samples were collected from six fields and interpolated via geographically weighted regression (GWR), random forest (RF), and inverse distance weighting (IDW). At four out of six site–years, the root mean square error of soil test P (STP) was 7 to 16% lower for the GWR method compared with the RF method, and GWR identified the low-STP areas better than RF. Geographically weighted regression may outperform RF and IDW because of its lower error and reduced sensitivity to individual high- and low-STP values. Evaluating multiple interpolation techniques and comparing maps both visually and using global error rates can improve the decisions made by farmers and other practitioners.