Comparison Of Management Zones Generated By The K-Means And Fuzzy C-Means Methods
The generation of Management Zones (MZ) is an economic alternative to make viable the precision agriculture (RODRIGUES & ZIMBACK, 2002) because they work as operation units for the inputs localized application and as soil and culture sample indicators. For the field division in areas with distinct characteristics, it is commonly used yield data, physical and chemical soil data, electrical conductivity, topography data and the combination between them (HORNUNG et al., 2006; FLEMING et al., 2004; KHOSLA et al., 2002). The clustering methods are highly suggested for the definition of MZ (TAYLOR et al., 2003; TAYLOR et al., 2007; YAN et al., 2007) Although there is the possibility of any attribute being related to the culture yield, for Doerge (2000), the ideal is to use temporally stable attributes correlated with yield. The most used clustering methods for defining management units correspond to the K-Means (FRIDGEN, et al., 2004; RIBEIRO, et al., 2011) and the Fuzzy C-Means algorithm (FRAISSE et al., 2001; STAFFORD et al., 1998; BOYDELL & MCBRATNEY, 2002; JAYMES et al., 2003; PING & DOBERMANN, 2003; YAN et al., 2007). They differ in the robustness added to the Fuzzy C-Means method, by Zadeh (1965), who introduced the theory of fuzzy logic to the division algorithm, enhanced by Ruspini (1969). This study aimed to evaluate if there is a difference between the clustering methods K-Means and Fuzzy C-Means. It has been done the selection of layers according Bazzi et al. (2013) and the evaluation was performed with the MZs mean comparison tests (ANOVA) and variance reduction. To evaluate the difference between MZs generated by the K-Means and Fuzzy C-Means methods, Kappa and Tau maps indices comparison were used. The layer altitude was selected as best option, were generated 2, 3, 4 and 5 MZs with both clustering methods. Comparing the thematic maps generated that represent the MZs, it was found that for two MZs, the divisions were the same (Kappa = 1 e Tau = 1). It may be classified as excellent agreement (LANDIS & KOCH (1977, p.165)) as well as the division into 3 classes (Kappa = 0,97 and TAU = 0,97). For the division into 4 classes, the agreement degree can be considered as substantial, and moderate for the division into 5 classes. It could be concluded that although there are differences between the methods used to generate MZs, the results were the same for both methods. It was found that the division is valid for both as to set different levels of yield in field and as to perform division of the field to use as a source of recommendation and analysis.
