Missing Data and Regression Models for Spatial Images

In previous work, we have shown that a functional concurrent linear model (FCLM) can be used to model the relationship between two spatial images. In this paper, we provide two extensions of the use of the FCLM to address missing data problems in series of colocated spatial images. First, we show how to build an FCLM relating two images involving gypsy moth defoliation data when there are missing data in some regions of the images. Because there is interest in filling in the missing scan lines in Landsat 7 images, we then further extend this approach to provide an imputation method for Landsat 7 data when the focus is on repairing a single image, rather than in relating images. A side effect of our approach is that the FCLM appears to automatically select the best parts of different covariate images for repairing a target image.