A Nonparametric Approach to the Removal of Documented Inhomogeneities in Climate Time Series

Abstract

limate data often suffer from artificial inhomogeneities, resulting from documented or undocumented events. For a time series to be used with confidence in climate analysis, it should only be characterized by variations intrinsic to the climate system. Many methods (e.g., direct or indirect) have been proposed according to the data characteristics (e.g., location, variable, or data completeness). This paper is focused on the abrupt-changes problem (when the properties of a time series change abruptly), when their timing is known, and suggests that a nonparametric regression framework provides an appealing way to correct for discontinuities in such a way as to recognize and allow for the existence of other structures such as seasonality and long-term smooth trends. The approach is illustrated by using reanalysis data for southern Africa, for which discontinuities are present because of the introduction of satellite technology in 1979.