Homogenization of extreme temperatures in Chile with GNSSfast

In Chile, in the last 20 years, significant changes have led to a notoriously warmer climate, which translates into meteorological records being recorded every year. For this reason, it is increasingly necessary for the National Meteorological Service to have reliable data to respond to the information needs of society and stakeholders.

Changes and modifications in meteorological stations, such as relocations, instruments, and environmental changes, can alter the estimation of means and long-term trends. Data homogenization consists of correcting these abrupt changes that are not associated with climate through different statistical methods to obtain reliable estimates. The most widely used homogenization methods are the relative ones, consisting of creating reference series from nearby stations, assuming they have the same long-term climatic signal.

In this work, the changing points of daily series of extreme temperatures were evaluated for 17 meteorological stations in Chile in the period 1961-2021. The segmentation method available in the R GNSSfast library was applied to the difference between a candidate series and a reference series created from neighboring stations. The correction of the inhomogeneous series was done through the difference in the daily median of the different segments, taking the most current period as a reference.

A total of 15 changepoints were detected in both minimum and maximum temperatures for 11 measurement stations, where 3 and 4 points were explained with dates of changes in the metadata, respectively. The most significant differences in the trend estimation with and without homogenization were observed in the weather stations of the country's central zone.