Combining Logistic Regression with Kriging for Mapping the Risk of Occurrence of Unexploded Ordnance (UXO)

1. Abstract This paper presents a methodology that combines logistic regression with kriging for incorporating exhaustive secondary information into the mapping of the risk of occurrence of unexploded ordnance (UXO). Logistic regression, which is appropriate for binary data (indicators) analysis, is used to derive the trend component in simple kriging with varying local means (SKlm). The technique is illustrated using two types of information: 1) hard indicators sampled along transects on a hypothetical UXO site generated using a doubly stochastic Poisson process, 2) exhaustive soft information obtained through the processing of a series of realizations generated by the same point process. After risks are mapped, pixels are flagged for further investigation if the estimated probability exceeds a given threshold. This classification is used to compare the performance of the proposed technique with traditional cokriging (collocated cokriging). Fewer misclassifications and smaller false positive rates are obtained for SKlm derived from logistic regression. The proportion of false negative is below 5% for both techniques.