Details of talk
|Title||Estimation of directional wind response using noisy bivariate circular data: a comparison of approaches|
|Presenter||Rachael Quill (The University of Adelaide)|
Observed bivariate datasets are considered noisy realisations of true underlying continuous signals to be estimated. Despite recent developments in toroidal kernel density estimation, there are limited alternatives to account for toroidality in nonparametric smoothing techniques. A simple and intuitive strategy to handle circular data is to wrap the dataset and reveal its circular nature prior to estimation. Traditional planar estimation techniques can then be utilised. This study aims to analyse this simple strategy in comparison to explicitly toroidal techniques. It is shown that using such an approach can produce as good, if not better surface estimation results than these toroidal methods. However, results are dependent upon the choice of planar estimation technique, and trade-offs between accuracy, smoothness and computational demand must be made in practice. Lessons from a simulation study are used in the estimation of bivariate wind direction distributions to understand the response of prevailing wind over mountainous terrain.