Data mining using k-means clustering and classification and regression trees (CART) as post-processing methods: identifying management and environmental factors for explaining sugarcane yield in Northern Argentina (1971-2005)

Final crop yield in an agroecosystem is determined by agronomic management andenvironmental factors interactions. Gains in understanding the magnitude and nature ofthese interactions are keys for the design of efficient and productive systems. Traditionally,this knowledge is acquired for a particular area indirectly, by means of simulation growthmodels (Lisson, et al., 2005). These models deliberately reduce the system complexity foridentifying the key aspects as well as to reduce the dataset required for parameterization.Alternatively, direct information of crop production can be stored in databases, whichdocument what has actually happened in the farming systems, capturing large scaleinformation on a wide range of variables that may potentially influence crop yield.However, the analysis of these large databases requires statistical methods capable ofdealing with multivariate and nonlinear data structures.