SPATIAL DEPENDENCES AND SAMPLING IN A FRUIT TREE - A NEW CONCEPT FOR SPATIAL PREDICTION IN FRUIT STUDIES

The complete description of a peach (Prunus persica (L.) Batsch) tree was used to estimate the spatial structure of dependence between fruits. Using a sample from the data, we check the actual behaviour of kriging for spatial prediction of fruit quality variables. Actual and predicted values and error characteristics were compared for spatial averages among small areas. The description of a second tree was used to simulate experimental situations with less data and to validate the extension of the dependence structure. Spatial sampling designs were compared for various size of predicted areas and for two prediction methods; kriging and a local unweighted mean. For each sampling design and method, the accuracy and precision of prediction were quantified with the mean-square error-of-prediction distribution (msep) using one hundred random sub-samples chosen among the exhaustive data set. The results show that kriging was a better predictor whatever the sampling design. When the sampling is not the best, or when the sampling rate decreases, kriging is less affected than the local unweighted mean. However, the gain in precision is sometimes too low to justify the use of kriging. Some common sampling designs among the tested ones, e.g. the nested designs, should not be used.