Revista Brasileira de Ciência do Solo
On-line version ISSN 1806-9657
ASSUMPCAO, Rosangela Aparecida Botinha; OPAZO, Miguel Angel Uribe and GALEA, Manuel. Local influence for spatial analysis of soil physical properties and soybean yield using student's t-distribution. Rev. Bras. Ciênc. Solo [online]. 2011, vol.35, n.6, pp.1917-1926. ISSN 1806-9657. http://dx.doi.org/10.1590/S0100-06832011000600008.
The modeling and estimation of the parameters that define the spatial dependence structure of a regionalized variable by geostatistical methods are fundamental, since these parameters, underlying the kriging of unsampled points, allow the construction of thematic maps. One or more atypical observations in the sample data can affect the estimation of these parameters. Thus, the assessment of the combined influence of these observations by the analysis of Local Influence is essential. The purpose of this paper was to propose local influence analysis methods for the regionalized variable, given that it has n-variate Student's t-distribution, and compare it with the analysis of local influence when the same regionalized variable has n-variate normal distribution. These local influence analysis methods were applied to soil physical properties and soybean yield data of an experiment carried out in a 56.68 ha commercial field in western Paraná, Brazil. Results showed that influential values are efficiently determined with n-variate Student's t-distribution.
Keywords : geostatistics; EM algorithm; spatial variability.