Assessing the Performance of the Gsimcli Homogenisation Method with Precipitation Monthly Data from the COST-HOME Benchmark

Abstract

Nowadays, climate data series are used in so many different studies that their importance implies the essential need of good data quality. For this reason, the process of homogenisation became a hot topic in the last decades, and many researchers have focused on developing efficient methods for the detection and correction of inhomogeneities in climate data series. This study evaluates the efficiency of the gsimcli homogenisation method, which is based on a geostatistical simulation approach. For each instant in time, gsimcli uses the direct sequential simulation algorithm to generate several equally probable realisations of the climate variable at the candidate station’s location, disregarding its values. The probability density function estimated at the candidate station’s location (local probability density functions (PDF)), for each instant in time, is then used to verify the existence of inhomogeneities in the candidate time series. When an inhomogeneity is detected, that value is replaced by a statistical value (correction parameter) derived from the estimated local PDF. In order to assess the gsimcli efficiency with different implementation strategies, we homogenised monthly precipitation data from an Austrian network of the COST-HOME benchmark data set (COST Action ES0601, Advances in homogenization methods of climate series: an integrated approach – HOME). The following parameters were tested: grid cell size, candidate order in the homogenisation process, local radius parameter, detection parameter and correction parameter. Performance metrics were computed to assess the efficiency of gsimcli. The results show the high influence of the grid cell size and of the correction parameter in the method’s performance.