Using the Wild Bootstrap Method on Wavelet-Decomposed Climate Time Series

• Activity Number: 342
• Type: Topic Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Quality and Productivity Section
• Author(s): Megan Heyman* and Snigdhansu Chatterjee and Amy Braverman and Noel Cressie
• Companies: University of Minnesota and University of Minnesota and Jet Propulsion Laboratory and University of Wollongong
• Keywords: bootstrap ; time series ; wavelets ; climate model ; resampling

Abstract

At a given grid cell location, climate models, such as the MIROC or IPSL, produce time series which claim to reflect the actual average climate conditions within that location. These output series may be considered as a single observation for phenomena like precipitation and temperature at a given time. Although we only have a single observation for a specified location, time, and phenomena, there is error associated with this observation. Understanding this error allows for model-to-model and model-to-observed climate comparisons, but estimating the error structure is difficult. A newly proposed resampling technique for time series is the wild bootstrap over wavelet coefficients. This method is flexible and produces consistent estimates of error and lag-correlations. We summarize the wild bootstrap method for wavelet coefficients, introduce an R package which performs the wavelet wild bootstrap, and present an application to climate model data.