Use this URL to cite or link to this record in EThOS:
Title: On smooth models for complex domains and distances
Author: Miller, David
ISNI:       0000 0004 2726 134X
Awarding Body: University of Bath
Current Institution: University of Bath
Date of Award: 2012
Availability of Full Text:
Access from EThOS:
Access from Institution:
Spline smoothing is a popular technique for creating maps of a spatial phenomenon. Most smoothers use the Euclidean metric to measure the distance between data. This approach is flawed since the distances between points in the domain as experienced by the objects within the domain are rarely Euclidean. For example, the movements of animals and people are subject to both physical and political boundaries (respectively) which must be navigated. Measuring distances between the objects using the incorrect (Euclidean) metric leads to incorrect inference. The first part of this thesis develops a finite area smoother which does not su↵er from this problem when the shape of the area is complex. It begins by rejecting the use of the Schwarz-Christo↵el transform as a method for morphing complex domains due to its squashing of space. From there a method based on preserving within-area distances using multidimensional scaling is developed. High dimensional projections of the data are necessary to avoid a loss of ordering in the points. To smooth reliably in high dimensions Duchon splines are used. The model developed rivals the current best finite area method in prediction error terms and fits easily into larger models. Finally, the utility of projection methods to smooth general distances is explored. The second part of the thesis concerns distance sampling, a widely used set of methods for estimating the abundance of biological populations. The work presented here introduces mixture formulation for the detection function used to model the probability of detection. The use of mixture models leads to flexible but monotonic detection functions, avoiding the unrealistic shapes which conventional methods are prone to. These new models are then applied to several existing, problematic data sets.
Supervisor: Wood, Simon Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: statistics ; smoothing ; distance sampling