How significantly different are your within field zones

Abstract

A classical approach in precision agriculture consists in validating within field zones defined from high spatial resolution observations by agronomic information (AI). Zones validation generally involves a two-step process. First, AI are obtained on a regular grid or following a target sampling strategy inside the field. Then, a statistical test, most often an ANOVA, is used to determine if the management zones created with the high spatial resolution auxiliary data explain differences in the AI values. Unfortunately, in precision agriculture, many of the works using such an approach omit a necessary condition for the implementation of the aforementioned ANOVA test, i.e. the observations need to be independent from each other. This condition is unfortunately seldom satisfied since AI are often spatially auto-correlated. In order to highlight this problem, simulated datasets with different and known AI spatial autocorrelation were used. Results show that as AI are more and more spatially auto-correlated, ANOVA tests almost always conclude that the management zones obtained with auxiliary data are significant whatever the zoning, i.e. even a completely random one. To overcome this problem, the paper introduces two methods directly inspired from published works in the field of ecology. Two cases were considered: the first one applies when large AI dataset (n>40) is available and the other one applies for small AI dataset (n<40). Both methods are implemented on a real precision viticulture example.