Bootstrapped high quantile estimation --- An experiment with scarce precipitation data

This is a Preprint and has not been peer reviewed. This is version 1 of this Preprint.

Add a Comment

You must log in to post a comment.


Comments

There are no comments or no comments have been made public for this article.

Downloads

Download Preprint

Supplementary Files
Authors

Hung Tan Thai Nguyen, Harald Bernhard, Zhangsheng Lai

Abstract

This paper details team SUTD’s effort when participating in the “Prediction of extremal precipitation” challenge. We propose a framework that combines the generalized Pareto distribution, a bootstrap resampling scheme and inverse distance weights to capture spatial dependence. Our method reduces the quantile loss functions by 55.1% as compared to a naive benchmark, and shows improvement across all months and all stations. The method works well even for stations without training data. Despite being simple, our method ranked fifth in the competition and our scores were very close to those of the winning teams. The framework is scalable and can be implemented easily by practising engineers.

DOI

https://doi.org/10.31223/osf.io/a6zj4

Subjects

Applied Statistics, Physical Sciences and Mathematics, Probability, Statistics and Probability

Keywords

generalized Pareto distribution, Ungauged sites, extreme precipitations, inverse distance weights

Dates

Published: 2018-02-25 15:39

License

Academic Free License (AFL) 3.0