relationships can be developed between spatial yield patterns for a given year and soil properties such as Crop yields are frequently heterogeneous across space and time. texture, apparent electrical conductivity, organic matter We performed this study to determine if cluster analysis could be used to decipher the temporal and spatial patterns of corn (Zea mays content, or terrain attributes (Yang et al., 1998). HowL .) yield within a field. Nonhierarchal cluster analysis was applied to ever, the relationships are not the same among years 6 yr of corn yield data collected for 224 yield plots on a regular grid (Jaynes et al., 1995b; Halvorson and Doll, 1991), unless on the southern half of a 32-ha field. We were able to group the yield the yield data is divided into subsets based on growing observations into five temporal yield patterns or clusters. The clusters season climatic conditions or field attributes (Kaspar et were not randomly distributed across the field but instead formed al., 2003; Timlin et al., 1998). For example, Kaspar et contiguous areas roughly equivalent to landscape positions. Cluster al. (2003) were able to develop a multiple-regression membership was determined primarily by yield differences in years equation based on terrain attributes that accounted for with growing season precipitation greater than the 40-yr average. A 78% of the yield variability for 6 yr of corn grown on multiple discriminant analysis was used to predict the spatial occura 16-ha field by restricting the analysis to the 4 yr having rence of the clusters from easily determined field attributes: soil electrical conductivity, elevation, slope, and plan and profile curvature. below-normal seasonal precipitation. When they in-The multiple discriminant functions were unable to distinguish be-cluded the 2 yr with above-normal growing season pretween the two clusters located on the lowest portions of the landscape. cipitation, only 26% of the yield variability could be Because of similar temporal yield patterns in these two clusters, they accounted for by multiple regression. were combined and the multiple discriminant analysis repeated for Rather than trying to predict specific yields within a four clusters. Using a holdout sample approach, we achieved 76 and field, we may be more successful in identifying areas 80% success rates in classifying the yield plots into the correct yield within a field that behave similarly among years. Most clusters. If response curves for inputs such as N prove to be unique studies have approached this problem by first identifor the different yield clusters, then clustering of multiple-year yield fying areas of similar soil properties or terrain attributes data may prove an effective method for determining management and then testing for reduction in yield variability or yield zones within fields. response to inputs within these areas (Fraisse et al., 2001; Fleming et al., 2000). These methods rely on using general knowledge of crop production to identify the