RESUMO:Este trabalho teve como objetivo comparar mapas temáticos construídos a partir de conjuntos de dados referentes à produtividade da soja, com diferentes grades amostrais regulares de 25x25 m; 50x50 m; 75x75 m e 100x100 m, utilizando técnicas de krigagem. No ajuste dos modelos teóricos a semivariâncias experimentais, utilizou-se para a estimação dos parâmetros o método de máxima verossimilhança. A comparação dos mapas temáticos foi realizada por meio dos índices de acurácia, obtidos a partir da matriz de erros. Foi verificado que fatores tais, como o tamanho amostral e a densidade amostral entre pontos, interferem na escolha do modelo teórico espacial, nas estimativas dos parâmetros e na construção dos mapas temáticos.PALAVRAS-CHAVE: dependência espacial, geoestatística, máxima verossimilhança. THEMATIC MAPS COMPARISON OF DIFFERENT SAMPLING GRIDS FOR SOYBEAN PRODUCTIVITYABSTRACT: The purpose of this paper was to compare thematic maps constructed from data sets related to soybean productivity with different regular sampling grids of 25x25 m, 50 x50 m, 75 x 75 m and 100x100 m, using kriging techniques. In the theoretical models fitted to the experimental semivariances, it was used the maximum likelihood method for parameter estimation. The comparison of thematic maps were made by accuracy indices, obtained from the error matrix. The results showed that factors such as sample size and sample density between points interfere in the choice of the theoretical spatial model, the parameter estimates and the construction of thematic maps.KEYWORDS: geostatistics, maximum likelihood, spatial dependence. INTRODUÇÃOCom o aumento da produção agrícola mundial e o uso de tecnologias para a mecanização, tornaram-se necessários o monitoramento e o gerenciamento do processo de produção agrícola, com o intuito de otimizá-los de forma racional. Nesse contexto, os métodos geoestatísticos vêm sendo utilizados no estudo da dependência espacial dos atributos físico-químicos do solo e da produtividade das culturas.Por meio desta análise, busca-se estimar os parâmetros que caracterizam a estrutura de dependência espacial, para posteriormente serem utilizados em técnicas de interpolação, como a krigagem, para fins de construção de mapas temáticos a serem considerados na tomada de decisão, com um melhor gerenciamento do processo produtivo das propriedades agrícolas (MOLIN, 2008;OLIVEIRA, et al., 2013;ASSUMPÇÃO et al., 2014).

Demand for quality weather forecasts has increased in the last decades, leading national meteorological centers to develop new forecasting models. These models have parameterizations which can produce different predictions for the same location and agrometeorological variable. In the state of Paraná - Brazil, studies on rain forecasting are important for planning the soybean crop. The objective of this study was to compare, based on a gold-standard and using bootstrapping residuals, forecasts of total rainfall by virtual stations of the following centers: Canadian Meteorological Center (CMC), European Center for Medium-Range Weather Forecasts (ECMWF), National Centers for Environmental Prediction (NCEP) and Center for Weather Forecasting and Climate Studies (CPTEC). Gold-standard measurements were obtained from Meteorological System of Paraná (SIMEPAR) meteorological stations. The studied region was the state of Paraná, in October–March of the harvest years 2011/2012–2015/2016; forecast ranges were 24 and 240 hours. Knowledge Discovery in Databases (KDD), focused on data mining techniques, was the chosen methodology. In the data preprocessing stage, spatial and temporal stratification, cleansing and grouping were performed. For the comparisons, 24 h and 240 h weather forecasts were used, being grouped in five-day and ten-day periods, respectively, and coefficients of agreement with the gold-standard measure were calculated. The choice of forecast center should consider the geographic location of a certain pluviometric station, and the temporal range of the forecast, according to its measure of agreement with the gold standard measure. Spatial variations of forecasting centers were identified within the mesoregions, which suggests the employment of different forecasting centers in a certain mesoregion.