Interpolation of Global Monthly Rain Gauge Observations for Climate Change Analysis

Abstract

ong-term global gridded datasets of observed precipitation are essential for the analysis of the global water and energy cycle, its variability, and possible changes. Several institutions provide those datasets. In 2005 the Global Precipitation Climatology Centre (GPCC) published the so-called Variability Analysis of Surface Climate Observations (VASClimO) dataset. This dataset is especially designed for the investigation of temporal change and variability. To date, however, the GPCC has not published how this dataset has been produced. This paper aims to fill this gap. It provides detailed information on how stations are selected and how data are quality controlled and interpolated. The dataset is based only on station records covering at least 90% of the period 1951?2000. The time series of 9343 stations were used. However, these stations are distributed very inhomogeneously around the globe; 4094 of these stations are within Germany and France. The VASClimO dataset is interpolated from relative deviations of observed monthly precipitation, leading to considerably lower interpolation errors than direct interpolation or the interpolation of absolute deviations. The retransformation from interpolated relative deviations to precipitation is done with local long-term averages of precipitation interpolated from data of the Food and Agriculture Organization of the United Nations. The VASClimO dataset has been interpolated with a method that is based on local station correlations (LSC) that is introduced here. It is compared with ordinary kriging and three versions of Shepard?s method. LSC outperforms these methods, especially with respect to the spatial maxima of interpolation errors.