Computational routines for the automatic selection of the best parameters used by interpolation methods to create thematic maps

Abstract

In precision agriculture, soil and plant variables are usually presented through thematic maps. The development of these maps is related to data collection, analysis, and interpolation. Although several mathematical interpolation methods are available (triangulation, natural-neighbor interpolation, inverse functions of distance, least-squares polynomials, and kriging), ordinary kriging (OK) and inverse distance weighting (IDW) are the most commonly used. However, farmers/agronomists are not highly trained in statistical methods to produce the best maps of soil and plant variables for precision agriculture. To ensure the best management approach is used, an automated method ready for application in an automated, easy-to-use mapping facility has the potential to be very useful. Thus, this study aimed to develop and apply computational routines capable of automatically identifying the best parameters for each interpolation method. Exhaustive tests were performed on these routines implemented in geoR software by using cross-validation methods and replacing the parameter values used by the interpolation techniques. The routines were applied to sample data encompassing corn and soybean yields as well as chemical and physical variables of soil collected in two agricultural areas located in the municipality of Serranópolis do Iguaçu, West Paraná, Brazil. For the semivariograms, 300 different adjustments were tested to identify the best parameters to interpolate the measured data using OK, and twelve different values were tested for the IDW exponent. As expected, the best results were obtained by OK when the variables exhibited spatial dependence. We concluded that the computational routines implemented are efficient and capable of identifying the interpolation method (OK with the spherical, exponential, Gaussian, and Matern family of semivariograms; IDW otherwise) with the best adjustment for each area as a function of the presence of spatial dependence.