MSc Stats Defence: A comparison of linear and non-linear kriging techniques for predicting the probability of exceeding a threshold value

CANDIDATE:   TIMOTHY DANCE

ABSTRACT:

This study investigates the effectiveness of different kriging techniques to predict the probability that the value of a variable at an unknown location exceeds a threshold value with particular application to variable rate application of fertilizer in agriculture. Most often disjunctive kriging is used to create probability maps of predictions, but it is shown here that simpler kriging techniques can be used to make predictions of similar quality when assumptions can be satisfied. Data was collected in 2017 using tractor mounted sensors on a farm in Southwestern Ontario. A measure of greenness, NDVI (the Normalized Difference Vegetative Index) was measured and linked to GPS coordinates. Both linear and non-linear methods show high predictive ability, suggesting that disjunctive kriging using Hermite polynomials is not necessary when the assumption of normality can be satisfied without non-linear transformations of the data. Additional data collection and analysis is suggested to validate these results and further refine the models.