Identifying Extreme Rainfall Events Using Functional Outliers Detection Methods

Mohanned Abduljabbar Hael,Yongsheng Yuan

doi:10.4236/jdaip.2020.84016

Abstract

Outlier detection techniques play a vital role in exploring unusual data of extreme events that have a critical effect considerably in the modeling and forecasting of functional data. The functional methods have an effective way of identifying outliers graphically, which might not be visible through the original data plot in classical analysis. This study’s main objective is to detect the extreme rainfall events using functional outliers detection methods depending on the depth and density functions. In order to identify the unusual events of rainfall variation over long time intervals, this work conducts based on the average monthly rainfall of the Taiz region from 1998 to 2019. Data were extracted from the Tropical Rainfall Measuring Mission and the analysis has been processed by R software. The approaches applied in this study involve rainbow plots, functional highest density region box-plot as well as functional bag-plot. According to the current results, the functional density box-plot method has proven effective in detecting outlier compared to the functional depth bag-plot method. In conclusion, the results of the current study showed that the rainfall over the Taiz region during the last two decades was influenced by the extreme events of years 1999, 2004, 2005, and 2009.

Highlights

Outlier detection approaches help in identifying characteristics that might have been neglected when using classical statistics and mathematical models
The approaches applied in this study involve rainbow plots, functional highest density region box-plot as well as functional bag-plot
The functional outlier detection methods performed in this study, functional (HDR) box-plot, made it easier to identify the extreme events of rainfall over the Taiz region and effectively detect unusual curves

Summary

Introduction

Outlier detection approaches help in identifying characteristics that might have been neglected when using classical statistics and mathematical models. This area of study has much attention when analyzing data in a functional context. Yuan [1] introduced the theoretical foundations and methodologies about functional data analysis with many applications besides [1] explained the characteristics of a functional form of a continuous variable over age or time. Many studies of several non-/parametric methods for the functional data analysis were extended by [2]. In the state-of-the-art literature on outlier detection in high dimensional data and functional data, various methods were presented by [3] [4] [5]

Objectives

Methods

Results

Conclusion