Spatial variability of wheat yield using the gaussian spatial linear model

Abstract

Wheat production has grown over the years and is one of the most important grain food sources for humans. This work analyzed the yield of two varieties of wheat planted in a regular sampling grid in an experimental area in the south of Brazil, using some explanatory variables. For the study of the spatial variability of wheat yield of the COODETEC 101 (CD101) and COODETEC 103 (CD103) varieties, which were cultivated by the farmer in an area of 22.62 ha, 84 samples were defined considering a regular grid of 50 x 50 m. In the sampled sites, the following explanatory variables were collected: average plant height in 60 days - avheight 60 (cm), the average number of tillers in 60 days - avtillers60 (cm), spike length in 120 days - splength (cm) and the wheat variety considered as a dummy variable (CD101 = 0 and CD103= 1). The wheat yield was analyzed using gaussian spatial linear models with different geostatistical models for the parametric form of the variance-covariance matrix. The significance of the parameters to select the explanatory variables were determined by the likelihood ratio test, and also a hypothesis test was presented to confirm that a model that deal with the spatial dependence was required by the data. To assess the global and local influence of some observations, diagnostics techniques based on Cook’s approach were considered. The disregard of potentially influential observations caused changes in the parameters estimates that define the spatial dependence structure, and consequently then in the profitability in sectors of the wheat yield maps. The study of statistical inference and diagnostics on spatial data should be part of all geostatistical analysis