SOME NEW DEVELOPMENTS ON TWO SEPARATE TOPICS: STATISTICAL CROSS VALIDATION AND FLOODPLAIN MAPPING

Abstract

This dissertation describes two unrelated threads of research. The first is a study of cross validation (CV), which is a data resampling method. CV is used for model ranking in model selection and for estimating expected prediction error of a model. A review of three resampling methods is provided in Chapter 1. Chapter 2 contains results from simulations that examine various properties of CV, in particular the use of CV for model selection in small sample settings as well as the expected value of the delete-d cross validation statistic. The second research thread is described in Chapter 3, where a new, physically-based computational model (called FLDPLN, or "Floodplain") for mapping potential inundation extents (floodplains) using gridded topographic data is introduced. Due to the parametric economy of FLDPLN, this model has significant advantages over existing methods such as hydrodynamic models. The model is validated using imagery from an actual flood event.