Functional Kriging Uncertainty Assessment

Geostatistical techniques for functional data were introduced by Goulard and Voltz (1993), but have only been developed recently. Several papers consider ordinary and universal kriging models to predict a curve at an unmonitored site under the assumption of a constant or longitude and latitude dependent mean or kriging with external drift, where scalar and functional exogenous variables are introduced. However, uncertainty evaluation of a predicted curve remains an open issue. Given the difficulty to derive sampling distributions for functional data, prediction band derivation can be approached using resampling methods. To evaluate uncertainty of a predicted curve, we adapt two semi-parametric bootstrap approach for spatially correlated data to the functional data case. The approach is illustrated by means of a simulation study.