Nonparametric Determination of Seasons

• Activity Number: 87
• Type: Invited
• Date/Time: Sunday, August 3, 2014 : 8:30 PM to 10:30 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Aaron Zimmerman*+ and Peter Guttorp
• Companies: University of Washington and University of Washington
• Keywords: nonparametric ; seasons ; nonstationary

Abstract

Statistical climate methods are often used to calculate seasonal estimates with the expectation that we observe different patterns or distributions of weather by season. Although an intuitively sensical approach, there are at least three different and oft-used definitions of 'seasons' (meteorological, astronomical, and rule-based), and there is no statistical reason to believe that seasonal climate patterns can be described by any one of these characterizations. Furthermore, performing a seasonal analysis within a (possible) climate change paradigm necessitates a method which allows for a non-stationary representation of the seasons. We propose a nonparametric method to partition the year into K seasons which are chosen as functions of the distribution of weather within each season. Using Kolmogorov-Smirnov statistics, we choose our seasonal boundaries to maximize the distributional differences between a season and the neighboring seasons. We demonstrate that even basic analyses are not robust to different definitions of seasons, and we apply the method to check for non-stationary seasonal bounds in a spatial network of temperature measurements (data from South-Central Sweden).