A Nonparametric and Functional Wombling Methodology

Abstract

Wombling methods, first introduced in 1951, have been widely applied to detect boundaries and variations across spatial domains, particularly in biological, public health and meteorological studies. Traditional applications focus on finite-dimensional observations, where significant changes in measurable traits indicate structural boundaries. In this work, wombling methodologies are extended to functional data, enabling the identification of spatial variation in infinite-dimensional settings. Proposed is a nonparametric approach that accommodates functional observations without imposing strict distributional assumptions. This methodology successfully captures complex spatial structures and discontinuities, demonstrating superior sensitivity and robustness compared to existing finite-dimensional techniques. This methodology is then applied to analyse regional epidemiological disparities between London and the rest of the UK, identifying key spatial boundaries in the shape of the first trajectory of Covid-19 incidence in 2020. Through extensive simulations and empirical validation, demonstrated is the method's effectiveness in uncovering meaningful spatial variations, with potential applications in a wide variety of fields.