Investigating the Potential of Geostatistical Methods for Defining Management Zones for Precision Farming of Cranberries at Two Scales

Characterization of within field variation in cranberry (Vaccinium macrocarpon Ait.) yield is difficult because the crop is harvested by flooding the bogs, agitating the vines and collecting the floating fruit. This gives one yield value per field. Remotely sensed images can give an indication of within field variation in cranberry yield, but noise in these need to be verified by intense and expensive ground surveys. Our objective was to infer within field variation of cranberry yield using a combination of remotely sensed images, yield data from about 700 beds in a region of southern NJ for a 14 year period (1991-2004) and geostatistical methods. Area-to-Area (AtoA) kriging was used to estimate yield values for cranberry fields that were not sampled in a given year by incorporating the size and shape of the fields in variogram deconvolution and kriging (regional scale). Area-to-Point kriging (AtoP) incorporated information on the size and shape of the fields (areal data) for making predictions to points on a 20 m grid (field scale). Principal components analysis was used to determine that mean growing season temperatures explained the patterns of yield over the 14 year period. AtoA and AtoP kriged surfaces of cranberry yield for hotter and colder than normal years were produced and combined with a vegetation index using non-hierarchical classification to stratify the region into management zones at a regional and field scale. In cold years the scale of variation was smaller and produced more classes both at the regional and field scale. When hotter and colder than average years were combined at the regional scale, three groups of fields were identified which largely corresponded to the three main cultivars of cranberry. The study highlights the importance of differences in sample support and how the scale of variation changes temporally with weather conditions.