The use of geostatistical tools combined with precision agriculture, allow the monitoring of agricultural soybean producing areas, establishing relationships of spatial dependence between the sampled points. The modeling of spatial variability structure enables the construction of thematic maps of the attributes studied, using the kriging as the interpolation method. However, the presence of outliers among the elements sampling can influence the construction and interpretation of these maps. The distribution of t-Student probability has been used in attempts to reduce the influence of atypical points in the estimation of parameters of spatial dependence, having heavier tails than the normal distribution of probabilities. The detection of influential points in the study area, through the analysis of local influence diagnostics, provides greater reliability in the use of maps, providing an efficient use of inputs. Then, the objective was to apply the techniques of local influence on spatially referenced data with models of disturbance and using the matrix additive scale, considering the distribution t-Student n-variate. It was used a linear spatial model for the study of soybean yield data as a function of average plant height and number of pods per plant. The local influence techniques were effective to detect points that influence in the geostatistical model selection, estimation of parameters and construction of thematic maps.
maximum likelihood; spatial variability; diagnostics
